A Method and System for Railway Edge Computing Service Caching and Task Offloading

Abstract

A method and system for railway edge computing service caching and task offloading. The method constructs an ACSCTO network model based on a pre-built railway system optimization model. In the ACSCTO network model, the service caching and task offloading network generates service caching decisions, and based on the service caching decisions, generates task offloading decisions. The bandwidth resource allocation network generates bandwidth resource allocation decisions based on the service caching and task offloading decisions. The computing resource allocation network generates computing resource allocation decisions based on the service caching, task offloading, and bandwidth resource allocation decisions. The ACSCTO network model is trained, and the trained ACSCTO network model is used to perform railway system service caching, task offloading, and resource allocation in real time. The ACSCTO network model proposed by the present invention can adaptively and dynamically adjust service caching and task offloading according to different task requirements, allocate computing resources and network bandwidth, thereby improving resource utilization rate, maximizing task processing efficiency, and reducing processing latency.

Description

Technical Field

[0001] The present invention belongs to the technical field of task offloading, and particularly relates to a method and system for caching and task offloading of railway edge computing services. Background Art

[0002] Mobile Edge Computing (MEC) significantly reduces the latency in data transmission by extending computing resources to locations close to railway vehicles (such as edge servers deployed along the route). MEC allows for real-time interaction and task processing in the railway system, thus minimizing network latency and improving the execution efficiency of tasks, especially in terms of time and energy consumption. Usually, the computing tasks of railway vehicles are offloaded to edge servers for processing. Since tasks such as train monitoring and passenger information services not only need to process user-specific data but also rely on large databases (such as passenger information databases or vehicle status databases), the transmission of these large-scale data may significantly increase communication latency, thereby affecting the overall performance of MEC. Moreover, the resources of edge servers are limited compared to cloud centers. In the railway scenario, the high-speed running trains and concurrent task processing of multiple trains further exacerbate the competition for bandwidth and computing resources. Tasks exhibit high heterogeneity in terms of data size, computing resource requirements, and latency constraints, which poses higher requirements for the optimal utilization of resources.

[0003] Existing technologies often focus on task offloading and service caching, without comprehensively considering the multi-dimensional optimization of the entire system. Although task offloading can reduce the processing latency of a single task, due to the neglect of the coordinated allocation of bandwidth and computing resources, it may lead to an overloaded resource load on the edge server, affecting the performance of the overall system. Furthermore, existing technologies usually focus on reducing task latency or improving resource utilization, often neglecting fairness among vehicles, especially when multiple railway vehicle tasks need to be processed. This may result in the preferential processing of computing tasks of certain vehicles, while the tasks of other vehicles are delayed, unable to meet the requirements of real-time and fairness in the railway transportation system. The resources of edge servers are limited, and how to fairly allocate these resources while ensuring system performance is a difficult point. In summary, existing technologies are often only applicable to scenarios with light loads. Due to the characteristics of high mobility, task heterogeneity, real-time requirements, and resource competitiveness of railway vehicles, the performance of existing technologies will drop sharply when facing high-density task requests in the railway system. Summary of the Invention

[0004] The object of the present invention is to address the above problems existing in the prior art, and provide a method and system for railway edge computing service caching and task offloading that adaptively adjusts service caching and task offloading, fairly allocates computing resources and network bandwidth, thereby improving the resource utilization rate of edge servers and railway vehicles locally, reducing processing latency and increasing the task completion rate, and can meet the task requirements of different types and complexities in railway scenarios.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] In the first aspect, the present invention provides a method for railway edge computing service caching and task offloading, the method comprising:

[0007] Step 1: Construct a railway system optimization model with the goal of maximizing average satisfaction;

[0008] Step 2: Construct an ACSCTO network model, the ACSCTO network model including a service caching and task offloading network, a bandwidth resource allocation network, and a computing resource allocation network, for first generating a service caching decision, and then generating a task offloading decision based on the generated service caching decision. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to generate a bandwidth resource allocation decision based on the service caching decision and the task offloading decision. The computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and is used to generate a computing resource allocation decision based on the service caching decision, the task offloading decision, and the bandwidth resource allocation decision;

[0009] Step 3: Train the ACSCTO network model;

[0010] Step 4: Use the trained ACSCTO network model to generate service caching decisions, task offloading decisions, and resource allocation decisions for the railway system, the resource allocation decisions including bandwidth resource allocation and computing resource allocation.

[0011] The training process of the ACSCTO network model includes:

[0012] S1. The service caching and task offloading network includes a service caching network and a task offloading network. The service caching network includes an Actor network and a Critic network. The Actor network and the Critic network have the same structure and share the same input data. Let the input data at time slot be , including at time slot The task information, service information, and cache status of the previous time slot, where the task information includes the number of services required by the task, the size of the task input data, the computing resources required by the task, and the channel gain between the task and the edge server, and the service information includes the cache status of the service in the edge server and the storage requirements; the output of the Actor network is transformed through a softmax layer to obtain a probability distribution vector , the probability distribution vector in which a single element represents the probability that the edge server caches the service in the time slot . Non-repetitive sampling is performed on the probability distribution vector to obtain a cache decision ; then the input data , the cache decision are combined to form the input data . The input data is input into the task offloading network, and multiple offloading decisions are output by the task offloading network;

[0013] S2. The cache decision , the offloading decision are input into the bandwidth resource allocation network together, and the bandwidth resource allocation decision is output by the bandwidth resource allocation module. The cache decision , the offloading decision , and the bandwidth resource allocation decision are input into the computing resource allocation network, and the computing resource allocation decision is output by the computing resource allocation network; based on the obtained cache decision , the offloading decision , the bandwidth resource allocation decision, and the computing resource allocation decision, the average satisfaction is calculated;

[0014] The offloading decision with the highest average satisfaction is selected and named as the offloading decision . The network parameters of the task offloading network are updated according to the following formula :

[0015] ;

[0016] In the above formula, represents the learning rate of the task offloading network; represents the gradient with respect to the task offloading network parameter ;

[0017] S3. Calculate the satisfaction of a single task in the offloading decision and name it as the reward . Calculate the satisfaction of a single task when all tasks are executed locally and name it as the reward . The offloading decision is calculated according to the following formula Offloading reward for a single task in : ;

[0018] S4. Combine to obtain a quadruple , where is the time slot of the input data; input the quadruple into the Critic network, and the Critic network outputs a vector of size , representing the number of edge server caching services, and the elements in the vector are , representing the value prediction value of selecting service for the input data ;

[0019] S5. Use to update the network parameters of the Critic network and the network parameters of the Actor network; the update formula is:

[0020] ;

[0021] ;

[0022] ;

[0023] In the above formula, represents the learning rate of the Critic network; represents the gradient with respect to the Critic network parameter ; represents the probability distribution of selecting an action according to the Actor network parameter given ; represents the target value of the input data ; represents the discount factor.

[0024] The objective function of the railway system optimization model is:

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] ; ;

[0030] ; ;

[0031] In the above formula, represents the average satisfaction; represents the computing resource vector; represents the service cache vector; represents the bandwidth allocation vector, represents the task offloading vector; represents task 's satisfaction; represents the scheduling period; represents the number of user devices; represents task 's maximum tolerable delay; represents task 's offloading decision, represents task executed locally, represents task offloaded to the edge server for execution; represents the computing resources allocated to task ; represents the bandwidth resources allocated to task in a single time slot; represents the satisfaction correction parameter; represents the overall completion time of the offloaded task; represents the total processing time of task executed locally; represents the service 's cache decision. When represents the service cached at the edge, when represents the service not cached; represents the wireless communication transmission delay of task from the local device to the edge server; represents the time required for service to be downloaded from the cloud to the edge server; represents the execution delay of task ; , respectively represent the cache status of the service required by task in time slot ; Represents the transmission rate between the edge server and the cloud; Represents the uplink data rate; Represents in the time slot Task The wireless channel gain between the local device and the edge server; Represents the transmission power of the local device; Represents the noise power spectral density; Represents the task The input data size of; Represents the service The storage requirement of; Represents the computing resources required for the task; Represents the task In the time slot Request service The indication variable of; Represents the average processing frequency of the user equipment;

[0032] The constraint conditions of the railway system optimization model include:

[0033] ;

[0034] ; ;

[0035] ; ;

[0036] ; ;

[0037] ;

[0038] ; ;

[0039] ; ;

[0040] In the above formula, Represents the total energy consumption of the edge server in the time slot ; Represents the energy consumed by the task in the time slot when executed on the edge server; Represents the maximum energy consumption allowed by the edge server; Represents in the time slot Task The energy consumed when transmitted on the edge server; Represents the total computing power of the edge server; represents the maximum cache capacity of the edge server; represents the number of services cached in the edge server; represents the total bandwidth of the edge server; represents the service 's average static power consumption; represents the effective switching capacitance coefficient related to the edge server.

[0041] The objective function of the bandwidth resource allocation model is: ;

[0042] ;

[0043] In the above formula, represents the satisfaction degree of the task after decomposition;

[0044] The constraint conditions of the bandwidth resource allocation model are: ;

[0045] ;

[0046] ;

[0047] ; ;

[0048] ;

[0049] ;

[0050] In the above formula, represents the Lambert W function; represents the minimum bandwidth resource required for the computing task under the condition of meeting its maximum tolerance delay; represents the maximum bandwidth resource required for the task when the service is not cached and needs to be downloaded from the cloud; and respectively represent the maximum and minimum bandwidth resources allocated to the task in a single time slot.

[0051] The objective function of the computing resource allocation model includes: ;

[0052] ;

[0053] The constraint conditions of the computing resource allocation model include:

[0054] Minimum resource constraint:

[0055] ;

[0056] In the above formula, represents the task in the time slot the minimum computing resources required;

[0057] Energy threshold constraint: ;

[0058] Computing resource upper limit constraint: ;

[0059] Introduce the Lagrangian function into the computing resource allocation model, and convert the computing resource allocation model into a convex problem for solution:

[0060] ;

[0061] In the above formula, is the Lagrangian function; , , respectively represent the Lagrange multipliers related to the minimum resource constraint, the energy threshold constraint, and the computing resource upper limit constraint.

[0062] In the second aspect, the present invention provides a railway edge computing service caching and task offloading system, and the system includes:

[0063] A railway system optimization model construction module, which is used to construct a railway system optimization model with the goal of maximizing the average satisfaction;

[0064] An ACSCTO network model construction module, which is used to construct an ACSCTO network model. The ACSCTO network model includes a service caching and task offloading network, a bandwidth resource allocation network, and a computing resource allocation network. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to first generate a service caching decision, and then generate a task offloading decision based on the generated service caching decision, and is used to generate a bandwidth resource allocation decision based on the service caching decision, the task offloading decision; the computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and generates a computing resource allocation decision based on the service caching decision, the task offloading decision, and the bandwidth resource allocation decision;

[0065] An ACSCTO network model training module, which is used to train the ACSCTO network model;

[0066] The actual operation module is used to obtain the service caching decision, task offloading decision, and resource allocation decision of the railway system by using the trained ACSCTO network model. The resource allocation decision includes bandwidth resource allocation and computing resource allocation.

[0067] The ACSCTO network model training module trains the ACSCTO model according to the following steps:

[0068] S1. The service caching and task offloading network includes a service caching network and a task offloading network. The service caching network includes an Actor network and a Critic network. The Actor network and the Critic network have the same structure and share the same input data. Let the input data of time slot be , including the task information, service information, and cache status of the previous time slot in time slot . The task information includes the number of services required by the task, the size of the task input data, the computing resources required by the task, and the channel gain between the task and the edge server. The service information includes the cache status and storage requirements of the service in the edge server. The output of the Actor network is transformed through a softmax layer to obtain a probability distribution vector . A single element in the probability distribution vector represents the probability that the edge server caches service in time slot . Non-repetitive sampling is performed on the probability distribution vector to obtain the caching decision . Then, the input data and the caching decision are combined to form the input data . The input data is input into the task offloading network, and multiple offloading decisions are output by the task offloading network;

[0069] S2. The caching decision and the offloading decision are input into the bandwidth resource allocation network together, and the bandwidth resource allocation decision is output by the bandwidth resource allocation module. The caching decision , the offloading decision , and the bandwidth resource allocation decision are input into the computing resource allocation network, and the computing resource allocation decision is output by the computing resource allocation network. Based on the obtained caching decision , the offloading decision , the bandwidth resource allocation decision, and the computing resource allocation decision, calculate the average satisfaction;

[0070] Select the offloading decision with the highest average satisfaction and name it the offloading decision , update the network parameters of the task offloading network according to the following formula :[[]]

[0071] ;

[0072] In the above formula, represents the learning rate of the task offloading network; represents the gradient with respect to the task offloading network parameter ;

[0073] S3. Calculate the satisfaction of a single task in and name it the reward , calculate the satisfaction of a single task when all tasks are executed locally and name it the reward , calculate the offloading reward of a single task in according to the following formula :[[]] ;

[0074] S4. Combine to obtain a quadruple , where is the input data at time slot ; input the quadruple into the Critic network, and the Critic network outputs a vector of size , represents the number of edge server caching services, and the elements in the vector are , represents the value prediction value of selecting service for the input data ;

[0075] S5. Use to update the network parameters of the Critic network and the network parameters of the Actor network; the update formula is:

[0076] ;

[0077] ;

[0078] ;

[0079] In the above formula, represents the learning rate of the Critic network; represents the gradient with respect to the Critic network parameter ; ​Represents the probability distribution of selecting an action according to the Actor network parameters under the given ; Represents the target value of the input data ; Represents the discount factor.

[0080] The objective function of the railway system optimization model is:

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] ; ;

[0086] ; ;

[0087] In the above formula, Represents the average satisfaction; Represents the computing resource vector; Represents the service cache vector; Represents the bandwidth allocation vector, Represents the task offloading vector; Represents the task satisfaction; Represents the scheduling period; Represents the number of user devices; Represents the task maximum tolerable delay; Represents the task offloading decision, Represents the task executed locally, Represents the task offloaded to the edge server for execution; Represents the computing resources allocated to the task ; Represents the bandwidth resources allocated to the task in a single time slot; Represents the satisfaction correction parameter; Represents the total completion time of the offloaded tasks; Represents the total processing time of the task executed locally; Represents the service The caching decision, when Indicates the service Is cached at the edge, when Indicates the service Is not cached; Indicates the task The wireless communication transmission delay from the local device to the edge server; Indicates the service The time required to download from the cloud to the edge server; Indicates the task The execution delay; 、 Respectively indicate the task In time slot 、Time slot The caching status of the required service; Indicates the transmission rate between the edge server and the cloud; Indicates the uplink data rate; Indicates in time slot Task The wireless channel gain between the local device and the edge server; Indicates the transmission power of the local device; Indicates the noise power spectral density; Indicates the task The input data size; Indicates the service The storage requirement; Indicates the computing resources required for the task; Indicates the task In time slot Request service The indicator variable; Indicates the average processing frequency of the user equipment;

[0088] The constraint conditions of the railway system optimization model include:

[0089] ;

[0090] ; ;

[0091] ; ;

[0092] ; ;

[0093] ;

[0094] ; ;

[0095] ; ;

[0096] In the above formula, represents the total energy consumption of the edge server in time slot ; represents the energy consumption for executing task on the edge server in time slot ; represents the maximum allowable energy consumption of the edge server; represents the energy consumption for transmitting task on the edge server in time slot ; represents the total computing power of the edge server; represents the maximum cache capacity of the edge server; represents the number of services cached in the edge server; represents the total bandwidth of the edge server; represents the average static power consumption of service ; represents the effective switching capacitance coefficient related to the edge server.

[0097] The objective function of the bandwidth resource allocation model is:

[0098] ;

[0099] ;

[0100] In the above formula, represents the satisfaction degree of task after decomposition;

[0101] The constraint conditions of the bandwidth resource allocation model are:

[0102] ;

[0103] ;

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] In the above formula, represents the Lambert W function; represents the computing task the minimum bandwidth resource required under the condition of satisfying its maximum tolerable delay; represents the maximum bandwidth resource required for the task when the service is not cached and needs to be downloaded from the cloud ; , and respectively represent the maximum and minimum bandwidth resources allocated to the task

[0110] The objective function of the computing resource allocation model includes: ;

[0111] ;

[0112] The constraint conditions of the computing resource allocation model include:

[0113] Minimum resource constraint: ;

[0114] In the above formula, represents the minimum computing resource required for the task in time slot ;

[0115] Energy threshold constraint: ;

[0116] Computing resource upper limit constraint: ;

[0117] Introduce the Lagrangian function into the computing resource allocation model to convert the computing resource allocation model into a convex problem for solution:

[0118] ;

[0119] In the above formula, is the Lagrangian function; , , respectively represent the Lagrange multipliers related to the minimum resource constraint, energy threshold constraint, and computing resource upper limit constraint.

[0120] Compared with the prior art, the beneficial effects of the present invention are:

[0121] 1. The railway edge computing service caching and task offloading method of the present invention constructs an ACSCTO model based on the Actor-Critic reinforcement learning algorithm. The ACSCTO model can adaptively and dynamically adjust service caching and task offloading according to the requirements of different tasks, allocate computing resources and network bandwidth, improve the resource utilization rate of edge servers and railway vehicle locals, maximize task processing efficiency, reduce processing latency and increase task completion rate. Especially in dealing with task requirements of different types and complexities in railway scenarios, such as real-time passenger information update, train status monitoring and intelligent scheduling, it can ensure effective processing of various tasks, reduce latency and improve overall operation efficiency.

[0122] 2. In the railway edge computing service caching and task offloading method of the present invention, when designing the satisfaction function, the fairness of resource allocation between trains is ensured through the concavity characteristic of the logarithmic function, preventing the situation where some trains occupy a large amount of computing or communication resources for a long time, which helps to maintain the overall stability of the railway system. BRIEF DESCRIPTION OF THE DRAWINGS

[0123] Figure 1 is a flowchart of the method of the present invention.

[0124] Figure 2 is a structural block diagram of the system of the present invention.

[0125] Figure 3 is a network architecture diagram of the ACSCTO network model of the present invention.

[0126] Figure 4 is a comparison chart of passenger satisfaction, average response time, task completion rate, fairness between vehicles, average overtime, energy consumption, average download latency, and average offloading quantity of different algorithms.

[0127] Figure 5 is a comparison chart of the convergence performance of different algorithms. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0128] The present invention will be further described in detail below in conjunction with the specific embodiments and the accompanying drawings.

[0129] Example 1:

[0130] Referring to Figure 1 , a railway edge computing service caching and task offloading method is carried out in the following steps in sequence:

[0131] Step 1: Construct an optimization model for the railway system with the goal of maximizing average satisfaction;

[0132] The railway system includes a cloud, an edge server (ES), and multiple train local devices (UD). Each local device generates at most one computing task in a single time slot , the computing task is indivisible and can be described as , where represents the size of the input data of the task, represents the computing resources required by the task, represents the maximum tolerable delay of the task; the computing tasks generated by local devices can be uploaded to the cloud. The edge server has caching and computing functions. The edge server is located next to the base station and can unload the tasks of local devices from the cloud to the edge server through the wireless network. The edge server determines which services should be cached and which tasks should be unloaded;

[0133] An optimization model of the railway system is constructed with the goal of maximizing the average satisfaction of tasks. When designing the average satisfaction, the maximum tolerable delay of the task is considered, and through the concavity property of the logarithmic function, the fair distribution of computing resources among local devices is achieved; finally, the objective function of the railway system optimization model is:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] ; ;

[0139] ; ;

[0140] In the above formula, represents the average satisfaction; is the bandwidth allocation vector, which represents the spectrum resources allocated to each task in a single time slot; is the computing resource vector, which represents the computing resources allocated to the tasks unloaded to the edge server in a single time slot; represents the service caching vector; represents the task offloading vector; represents the task 's satisfaction; represents the scheduling period; represents the number of user devices; represents the task 's maximum tolerable delay; represents the task 's offloading decision, represents the task Execute locally, indicating the task is offloaded to the edge server for execution; indicating the computing resources allocated to the task ; indicating the bandwidth resources allocated to the task in a single time slot; indicating the satisfaction correction parameter; indicating the overall completion time of the offloaded task; indicating the task total processing time executed on the local device; indicating the service cache decision, when indicating the service is cached at the edge, when indicating the service is not cached; indicating the task wireless communication transmission delay from the local device to the edge server; indicating the service time required to download from the cloud to the edge server; indicating the task execution delay; , respectively indicating the task in time slot , time slot cache status of the required service; indicating the transmission rate between the edge server and the cloud; indicating the uplink data rate; indicating in time slot the task wireless channel gain between the local device and the edge server; indicating the transmission power of the local device; indicating the noise power spectral density; indicating the task input data size; indicating the service storage requirement; indicating the computing resources required by the task; indicating the task in time slot requesting the service indicator variable; indicating the average processing frequency of the user equipment;

[0141] The constraint conditions of the railway system optimization model include:

[0142] ;

[0143] ; ;

[0144] ; ;

[0145] ; ;

[0146] ;

[0147] ; ;

[0148] ; ;

[0149] In the above formula, represents the total energy consumption of the edge server at time slot ; represents the energy consumption for executing task on the edge server at time slot ; represents the maximum energy consumption allowed for the edge server; represents the energy consumption for transmitting task on the edge server at time slot ; represents the total computing power of the edge server; represents the maximum cache capacity of the edge server; represents the number of services cached in the edge server; represents the total bandwidth of the edge server; represents the average static power consumption of service ; represents the effective switching capacitance coefficient related to the edge server;

[0150] To simplify the railway system optimization model, it is divided into two sub - problems: service caching and task offloading problem, and resource allocation problem (MRA); the MRA problem can also be decomposed into two sub - problems: bandwidth resource allocation problem (BRA) and computing resource allocation problem (CRA). The BRA problem and the CRA problem can be solved by an alternating iteration method until the problem converges; for the cache decision, offloading decision, and bandwidth resource allocation decision given for a single time slot obtained previously, the following bandwidth resource allocation model is constructed:

[0151] The objective function of the bandwidth resource allocation model is:

[0152] ;

[0153] ;

[0154] In the above formula, represents the satisfaction degree of the task after deconstruction;

[0155] The constraint conditions of the bandwidth resource allocation model are: ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] In the above formula, represents the Lambert W function; represents the minimum bandwidth resource required for the computing task under the condition of meeting its maximum tolerance delay; represents the maximum bandwidth resource required for the task when the computing service is not cached and needs to be downloaded from the cloud; , respectively represent the maximum and minimum bandwidth resources allocated to the task in a single time slot;

[0163] Introduce the Lagrangian function into the bandwidth resource allocation model; the introduced Lagrangian function can be expressed as:

[0164] ;

[0165] In the above formula, is the Lagrangian function introduced into the bandwidth resource allocation model; is the Lagrange multiplier vector, and each multiplier corresponds to a constraint condition in the bandwidth resource allocation model;

[0166] Calculate according to the following formula and update using the bisection method until converges;

[0167] ;

[0168] ; ; ;

[0169] For the cache decision, offloading decision, and bandwidth allocation decision obtained previously, the following computing resource allocation model is constructed:

[0170] The objective function of the computing resource allocation model includes:

[0171] ;

[0172] ;

[0173] The constraint conditions of the computing resource allocation model include:

[0174] The minimum resource constraint is used to ensure that the computing resources allocated to each task are not less than the minimum resource amount required:

[0175] ;

[0176] In the above formula, represents the minimum computing resource required for task in time slot ;

[0177] The energy threshold constraint is used to ensure that the total energy consumption of the entire network does not exceed a predetermined threshold:

[0178] ;

[0179] The computing resource upper limit constraint is used to ensure that the total amount of computing resources allocated to all tasks does not exceed the total computing power of the edge server:

[0180] ;

[0181] Since the computing resource allocation model has linear and convex problems, the Lagrangian function is introduced into the computing resource allocation model to convert the computing resource allocation model into a convex problem for solution; the Lagrangian function can be expressed as:

[0182] ;

[0183] In the above formula, is the Lagrangian function; , , respectively represent the Lagrange multipliers related to the minimum resource constraint, energy threshold constraint, and computing resource upper limit constraint;

[0184] After determining the Lagrange multipliers, calculate by solving the following function:

[0185] ;

[0186] Iteratively update the Lagrange multipliers until is satisfied, where and are the computing resources allocated to task in the iteration and the iteration respectively, is the tolerance of ; the iterative update formula for the Lagrange multipliers is as follows:

[0187] ;

[0188] ;

[0189] ;

[0190] In the above formula, , , all represent positive gradient step sizes; represents the current iteration number; , are the Lagrange multipliers obtained in the iteration and the iteration respectively; , are the Lagrange multipliers obtained in the iteration and the iteration respectively; , are the Lagrange multipliers obtained in the iteration and the iteration respectively;

[0191] Approximate the optimal solution by continuously updating the Lagrange multipliers. During the process of approximating the optimal solution, the computing resource allocation decision can be dynamically adjusted to adapt to the changes in user requirements and network conditions, thereby realizing dynamic computing resource allocation;

[0192] Step 2: Construct the ACSCTO network model, as Figure 3As shown in the figure, the ACSCTO network model includes a service cache and task offloading network, a bandwidth resource allocation network, and a computing resource allocation network, which are used to first generate a service cache decision, and then generate a task offloading decision based on the generated service cache decision. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to generate a bandwidth resource allocation decision according to the service cache decision and the task offloading decision. The computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and is used to generate a computing resource allocation decision according to the service cache decision, the task offloading decision, and the bandwidth resource allocation decision;

[0193] Step 3: Train the ACSCTO network model; as Figure 3 shown, the training process of the ACSCTO network model includes the following steps:

[0194] S1. The ACSCTO network model makes service cache and task offloading decisions based on the Actor-Critic framework; the service cache and task offloading network includes a service cache network and a task offloading network. The service cache network includes an Actor network and a Critic network. The structures of the Actor network and the Critic network are the same and share the same input data. Let the input data at time slot be , which includes task information, service information, and the cache status of the previous time slot at time slot . The task information includes the number of services required by the task, the size of the task input data, the computing resources required by the task, and the channel gain between the task and the edge server. The service information includes the cache status, energy consumption, and storage requirements of the service on the edge server. The output of the Actor network is transformed through a softmax layer to obtain a probability distribution vector . A single element in the probability distribution vector represents the probability that the edge server caches service at time slot . Each probability value is between 0 and 1, and the sum of the probability distribution is equal to 1. Non-repetitive sampling is performed on the probability distribution vector to obtain a cache decision . Then, the input data , the cache decision are combined to form input data . The input data is input into the task offloading network, and multiple offloading decisions are output by the task offloading network;

[0195] S2. The cache decision , the offloading decision An input bandwidth resource allocation network outputs a bandwidth resource allocation decision by a bandwidth resource allocation module, and combines the caching decision , offloading decision , and bandwidth resource allocation decision and inputs them into a computing resource allocation network, which outputs a computing resource allocation decision; based on the obtained caching decision , offloading decision , bandwidth resource allocation decision, and computing resource allocation decision, calculate the average satisfaction degree;

[0196] For each time slot, select the offloading decision with the highest average satisfaction degree and name it the offloading decision , and combine the offloading decision and its corresponding input data into a binary tuple and store it in a buffer with limited capacity . When the buffer memory reaches the full capacity, the newly generated binary tuple will replace the oldest binary tuple in the buffer . Randomly select a batch of binary tuples from the buffer and update the network parameters of the task offloading network according to the following formula :

[0197] ;

[0198] In the above formula, represents the learning rate of the task offloading network; represents the gradient with respect to the task offloading network parameter ;

[0199] S3. Calculate the satisfaction degree of a single task in the offloading decision and name it the reward , calculate the satisfaction degree of a single task when all tasks are executed locally and name it the reward , and calculate the offloading reward of a single task in the offloading decision according to the following formula

[0200] ;

[0201] S4. Generate a quadruple for each time slot , where is the input data for time slot ; store the newly generated quadruple at each time step in a memory with limited capacity . When the memory memory reaches the full capacity, the newly generated quadruple will replace the quadruple in the memory The oldest quadruple; randomly extract a batch of quadruples from the memory and input them into the Critic network. The Critic network generates a state prediction value for each service in the current time slot. The output size of the Critic network is vector , denotes the number of services cached by the edge server. The elements in the vector are , denotes the value prediction value of the selected service for the input data; it should be noted that the in should be understood as the service ;

[0202] S5. Use to update the network parameters of the Critic network and the network parameters

[0203] of the Actor network; the update formula is:

[0204] ;

[0205] ;

[0206] In the above formula, denotes the learning rate of the Critic network; denotes the gradient with respect to the Critic network parameter ; denotes the probability distribution of selecting an action according to the Actor network parameter given ; denotes the target value of the input data ; denotes the discount factor;

[0207] Adopt the backpropagation algorithm and the offloading reward derived from the task offloading network to optimize the network parameters of the Actor network and the Critic network. The network parameters generally refer to weights and biases; by iteratively adjusting the network parameters, the performance of both the Actor network and the Critic network is gradually enhanced, so as to make more accurate decisions during the service caching process;

[0208] The specific training process of the ACSCTO network model is as follows:

[0209]

[0210] Step 4: Use the trained ACSCTO network model to generate service caching decisions, task offloading decisions, and resource allocation decisions for the railway system. The resource allocation decisions include bandwidth resource allocation and computing resource allocation.

[0211] To determine the optimal scheduling decision for each time slot, on the one hand, the ACSCTO model of the present invention jointly optimizes service caching and task offloading to improve the overall performance. By iteratively updating the service caching network and the task offloading network, and using the rewards derived from the task offloading model to optimize the service caching network, the coordination and consistency between the caching decision and the offloading decision can be ensured, thereby maximizing the satisfaction. On the other hand, based on the caching decision and the offloading decision obtained from the service caching network and the task offloading network, the resource allocation network is used to determine the optimal resource allocation strategy, so that the ACSCTO model can dynamically adapt to the environment and more effectively solve complex tasks and service requirements.

[0212] Performance verification:

[0213] To verify the effectiveness of the method of the present invention, a railway model simulation model is constructed. The ACSCTO algorithm of the present invention and other baseline algorithms are used for simulation calculations on the railway model simulation model. The simulation calculations are carried out for 10,000 time slots, and the performance of each algorithm is compared according to the simulation results and the pre-constructed evaluation index system.

[0214] In the railway model simulation model, the edge server is located at the railway communication base station, covering an area of 100 km × 100 km. 30 trains are evenly distributed in this area. To meet the high real-time requirements of the trains, the storage capacity of the edge server is set to 50 GB, the computing power is 80 GHz, and the total network capacity is 60 GHz to ensure that it can handle task requests sent by multiple trains simultaneously. The railway model simulation model provides ten train services, covering key tasks such as train status monitoring, real-time passenger information update, and train scheduling. Each service occupies storage resources in the range of about 2 GB to 8 GB, and the energy consumption required for each service cache is about 3 to 7 units. The input data size of the computing tasks of each railway vehicle is about 5 MB to 20 MB, the computing resources required for the tasks are about 2 GHz to 10 GHz, and the task execution period is 1 second to 5 seconds.

[0215] Other baseline algorithms include: AC uniform allocation method (AC-ea), AC single value method (AC-sr), non-decomposed AC method (AC-wd), popularity caching method (PC), local computing method (LP); the AC uniform allocation method adopts the same caching and offloading strategies as the ACSCTO algorithm, but uniformly allocates bandwidth and computing resources (computing resources are the CPU frequency) to offloaded tasks; the difference between the AC single value method and the ACSCTO algorithm is that it directly calculates the state estimation value of the entire offloading decision using the A2C framework, rather than calculating the state estimation value based on a single service; the difference between the non-decomposed AC method and the ACSCTO algorithm is that it directly determines the caching decision and offloading decision after obtaining the state estimation value of a single service; the popularity caching method makes caching decisions based on the popularity of services, where the popularity of services is defined according to the frequency of service requests, and tasks are prioritized according to the urgency of tasks (i.e., task delay constraints), and tasks with stricter deadlines will be assigned higher priorities; the local computing method is that all tasks are executed locally.

[0216] The evaluation index system includes: convergence performance, passenger satisfaction, average response time, task completion rate, fairness among vehicles, average overtime, energy consumption, average download delay, average number of offloads. The calculation results of each evaluation index are as follows:

[0217] (1) The convergence curve obtained by simulation is as Figure 5 shown. From Figure 5 it can be seen that these algorithms show rapid convergence in the initial stage when the time frame is less than 5000, and then gradually decelerate, approaching convergence at about 20,000 time frames. Although these algorithms show similar convergence speeds, their final rewards are significantly different. The proposed ACSCTO algorithm of the present invention shows the highest final reward, which verifies the effectiveness of the method proposed in the present invention. The final reward of the AC uniform allocation method exceeds that of the AC single value method, which highlights the advantage of calculating the state estimation value based on a single service in optimizing task scheduling and resource allocation.

[0218] (2) The calculated passenger satisfaction, average response time, task completion rate, fairness among vehicles, average overtime, energy consumption, average download delay, average number of offloads are as Figure 4 shown in (a)-(h) in Figure 4As can be seen from Figure (a), the passenger satisfaction of all algorithms is negative. Among the five baseline algorithms, the AC uniform allocation method shows higher user satisfaction. Compared with the AC single-value method, the AC uniform allocation method allocates less time for service update, but has a higher task offloading rate to the edge server. This indicates that the AC uniform allocation method has the ability to optimize service caching decisions, thereby reducing service download latency, improving task offloading decisions, and enhancing user satisfaction. However, the passenger satisfaction of the proposed ACSCTO algorithm of the present invention exceeds that of the best-performing AC uniform allocation method among all baseline algorithms because the ACSCTO algorithm not only inherits all the advantages of the AC uniform allocation method but also integrates a Lagrangian function convex optimizer for resource allocation, improving the upper limit of the overall algorithm. Therefore, ACSCTO shows a rapid adaptation to vehicle tasks and provides greater flexibility in resource allocation. When faced with urgent offloading tasks, the ACSCTO model will allocate more bandwidth and computing resources to prevent the delay from exceeding the time limit, thereby avoiding the penalty caused by the delayed completion of tasks. Therefore, as Figure 4 shown in Figures (b)-(e), the ACSCTO algorithm has the shortest average response time, the lowest average overtime, the highest task completion rate, and vehicle-to-vehicle fairness. At the same time, the energy consumption of the ACSCTO algorithm is still comparable to that of the AC uniform allocation method, which means that the ACSCTO algorithm can achieve the preferential utilization of limited resources. Since the non-decomposed AC method does not separate service caching from task offloading but uses the A2C framework to determine the optimal caching and offloading decisions, it cannot understand the correspondence between services and tasks and lacks the consistency between service caching and task offloading, resulting in a significant reduction in the final reward. The popularity caching method is a heuristic algorithm designed to cache frequently requested services on the edge server and preferentially offload more urgent tasks to the edge server. However, it lacks generality and rapid adaptability. It not only shows the second-worst performance after the local computing method in terms of passenger satisfaction, average response time, and task completion rate but also has the worst user-to-user fairness and significant fluctuations in the average offloading quantity.

[0219] Embodiment 2:

[0220] See Figure 2, a railway edge computing service caching and task offloading system, the system includes a railway system optimization model construction module, an ACSCTO network model construction module, an ACSCTO network model training module, and an actual operation module. The railway system optimization model construction module is used to construct a railway system optimization model with the goal of maximizing the average satisfaction. The objective function and constraints of the railway system optimization model are as shown in Step 1 of Embodiment 1. The ACSCTO network model construction module is used to construct an ACSCTO network model, which includes a service caching and task offloading network, a bandwidth resource allocation network, and a computing resource allocation network. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to first generate a service caching decision, and then generate a task offloading decision based on the generated service caching decision. The computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and generates a computing resource allocation decision based on the service caching decision, the task offloading decision, and the bandwidth resource allocation decision. The objective function and constraints of the bandwidth resource allocation model and the objective function and constraints of the computing resource allocation model are both as shown in Step 1 of Embodiment 1. The ACSCTO network model training module is used to train the ACSCTO network model. The specific training steps of the ACSCTO network model are as shown in Step 3 of Embodiment 1. The actual operation module is used to obtain the service caching decision, the task offloading decision, and the resource allocation decision of the railway system by using the trained ACSCTO network model, and the resource allocation decision includes bandwidth resource allocation and computing resource allocation.

Claims

1. A method for caching and task offloading in railway edge computing services, characterized in that: The method includes: Step 1: Optimize service caching decisions, task offloading decisions, bandwidth resource allocation decisions, and computing resource allocation decisions with the goal of maximizing average satisfaction, and construct a railway system optimization model; the average satisfaction is calculated based on the maximum tolerable delay of tasks, satisfaction correction parameters, the overall completion time of offloaded tasks, and the total processing time of tasks executed locally; Step 2: Construct an ACSCTO network model, which includes a service caching and task offloading network, a bandwidth resource allocation network, and a computing resource allocation network, used to first generate service caching decisions, and then generate task offloading decisions based on the generated service caching decisions. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to generate bandwidth resource allocation decisions based on service caching decisions and task offloading decisions. The computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and is used to generate computing resource allocation decisions based on service caching decisions, task offloading decisions, and bandwidth resource allocation decisions; Step 3: Train the ACSCTO network model; Step 4: Use the trained ACSCTO network model to generate service caching decisions, task offloading decisions, and resource allocation decisions for the railway system. The resource allocation decisions include bandwidth resource allocation and computing resource allocation; The training process of the ACSCTO network model includes: S1. The service cache and task offloading network includes a service cache network and a task offloading network. The service cache network includes an Actor network and a Critic network. The Actor network and the Critic network have the same structure and share the same input data. Let the time slot have an input data of , which includes the task information, service information, and cache status of the previous time slot in the time slot . The task information includes the number of services required for the task, the size of the task input data, the computing resources required for the task, and the channel gain between the task and the edge server. The service information includes the cache status and storage requirements of the service in the edge server. The output of the Actor network is transformed by a softmax layer to obtain a probability distribution vector . A single element in the probability distribution vector represents the probability that the edge server caches the service in the time slot . Non-repetitive sampling is performed on the probability distribution vector to obtain a cache decision . Then, the input data and the cache decision are combined to form an input data . The input data is input into the task offloading network, and the task offloading network outputs multiple offloading decisions . S2. Input the caching decision and the offloading decision into the bandwidth resource allocation network together. The bandwidth resource allocation module outputs the bandwidth resource allocation decision. Input the caching decision and the offloading decision as well as the bandwidth resource allocation decision into the computing resource allocation network. The computing resource allocation network outputs the computing resource allocation decision. Based on the obtained caching decision , the offloading decision , the bandwidth resource allocation decision, and the computing resource allocation decision, calculate the average satisfaction degree; Select the offloading decision with the highest average satisfaction and name it the offloading decision , and update the network parameters of the task offloading network according to the following formula :[[]]END]] ； In the above formula, represents the learning rate of the task offloading network; represents the gradient with respect to the task offloading network parameter ; S3. Compute offloading decision The satisfaction of a single task in and name it as the reward. Compute the satisfaction of a single task when all tasks are executed locally and name it as the reward According to the following formula, the offloading decision is obtained The offloading reward of a single task in : ； S4. Combine to obtain a quadruple , where is the input data for time slot . Input the quadruple into the Critic network, and the Critic network outputs a vector of size , indicating the number of edge server caching services. The elements in the vector are , indicating the value prediction value of selecting service for the input data . S5. Use to update the network parameters of the Critic network and the network parameters of the Actor network The update formula is as follows: ； ； ； In the above formula, represents the learning rate of the Critic network; represents the gradient with respect to the Critic network parameter ; represents the probability distribution of selecting an action according to the Actor network parameter under the given ; represents the target value of the input data ; represents the discount factor.

2. A method for caching and task offloading in railway edge computing services according to claim 1, characterized in that: The objective function of the railway system optimization model is: ； ； ； ； ； ； ； ； In the above formula, represents the average satisfaction; represents the computing resource vector; represents the service cache vector; represents the bandwidth allocation vector, represents the task offloading vector; represents the task satisfaction; represents the scheduling period; represents the number of user devices; represents the task maximum tolerable delay; represents the task offloading decision, represents the task executed locally, represents the task offloaded to the edge server for execution; represents the computing resource allocated to the task ; represents the bandwidth resource allocated to the task in a single time slot; represents the satisfaction correction parameter; represents the overall completion time of the offloaded task; represents the total processing time of the task executed locally; represents the service cache decision, when represents the service cached at the edge, when represents the service not cached; represents the wireless communication transmission delay of the task from the local device to the edge server; represents the time required for the service to be downloaded from the cloud to the edge server; represents the execution delay of the task ; 、 respectively represent the cache status of the service required by the task in time slot 、 time slot ; represents the transmission rate between the edge server and the cloud; represents the uplink data rate; represents the wireless channel gain between the local device and the edge server for the task in time slot ; represents the transmit power of the local device; represents the noise power spectral density; represents the task of the input data size; represents the service of the storage requirement; represents the computing resources required for the task; represents the task in the time slot requests the service of the indicator variable; represents the average processing frequency of the user equipment; The constraint conditions of the railway system optimization model include: ； ； ； ； ； ； ； ； ； ； ； ； In the above formula, represents the total energy consumption of the edge server in time slot ; represents the energy consumption for executing task on the edge server in time slot ; represents the maximum allowable energy consumption of the edge server; represents the energy consumption for transmitting task on the edge server in time slot ; represents the total computing power of the edge server; represents the maximum cache capacity of the edge server; represents the number of services cached in the edge server; represents the total bandwidth of the edge server; represents the average static power consumption of service ; represents the effective switching capacitance coefficient related to the edge server.

3. A method for caching and task offloading in railway edge computing services according to claim 2, characterized in that: The objective function of the bandwidth resource allocation model is: ； ； In the above formula, represents the task in the satisfaction after deconstruction; The constraint conditions of the bandwidth resource allocation model are: ； ； ； ； ； ； ； In the above formula, represents the Lambert W function; represents the computing task the minimum bandwidth resource required under the condition of satisfying its maximum tolerance delay; represents the maximum bandwidth resource required for the task when the service is not cached and needs to be downloaded from the cloud ; and respectively represent the maximum and minimum bandwidth resources allocated to the task in a single time slot.

4. A method for caching and task offloading in railway edge computing services according to claim 2 or 3, characterized in that: The objective function of the computing resource allocation model includes: ； ； The constraint conditions of the computing resource allocation model include: Minimum resource constraint: ； In the above formula, represents the task in the time slot required minimum computing resources; Energy threshold constraint: ； Computing resource upper limit constraint: ； Introduce the Lagrangian function into the computing resource allocation model to convert the computing resource allocation model into a convex problem for solution: ； In the above formula, is the Lagrangian function; , , represent the Lagrange multipliers related to the minimum resource constraint, the energy threshold constraint, and the upper limit constraint of computing resources, respectively.

5. A railway edge computing service caching and task offloading system, characterized in that: The system includes: A railway system optimization model construction module, used to optimize service caching decisions, task offloading decisions, bandwidth resource allocation decisions, and computing resource allocation decisions with the goal of maximizing average satisfaction, and construct a railway system optimization model; the average satisfaction is calculated based on the maximum tolerable delay of tasks, satisfaction correction parameters, the overall completion time of offloaded tasks, and the total processing time of tasks executed locally; The ACSCTO network model construction module is used to construct the ACSCTO network model. The ACSCTO network model includes a service cache and a task offloading network, a bandwidth resource allocation network, and a computing resource allocation network. The bandwidth resource allocation network is constructed based on the bandwidth resource allocation model deconstructed from the railway system optimization model, and is used to first generate a service cache decision, and then generate a task offloading decision based on the generated service cache decision, and is used to generate a bandwidth resource allocation decision based on the service cache decision and the task offloading decision. The computing resource allocation network is constructed based on the computing resource allocation model deconstructed from the railway system optimization model, and generates a computing resource allocation decision based on the service cache decision, the task offloading decision, and the bandwidth resource allocation decision. The ACSCTO network model training module is used to train the ACSCTO network model. The actual operation module is used to obtain the service cache decision, the task offloading decision, and the resource allocation decision of the railway system by using the trained ACSCTO network model. The resource allocation decision includes bandwidth resource allocation and computing resource allocation. The ACSCTO network model training module trains the ACSCTO network model according to the following steps: S1. The service cache and task offloading network includes a service cache network and a task offloading network. The service cache network includes an Actor network and a Critic network. The Actor network and the Critic network have the same structure and share the same input data. Let the time slot input data be , including the task information, service information, and cache status of the previous time slot in the time slot . The task information includes the number of services required for the task, the size of the task input data, the computing resources required for the task, and the channel gain between the task and the edge server. The service information includes the cache status and storage requirements of the service in the edge server. The output of the Actor network is transformed by a softmax layer to obtain a probability distribution vector . A single element in the probability distribution vector represents the probability that the edge server caches the service in the time slot . Non-repetitive sampling is performed on the probability distribution vector to obtain a cache decision . Then, the input data and the cache decision are combined to form input data . The input data is input into the task offloading network, and the task offloading network outputs multiple offloading decisions . S2. Input the caching decision and the offloading decision into the bandwidth resource allocation network together. The bandwidth resource allocation module outputs the bandwidth resource allocation decision. Input the caching decision and the offloading decision and the bandwidth resource allocation decision into the computing resource allocation network. The computing resource allocation network outputs the computing resource allocation decision. Based on the obtained caching decision and the offloading decision and the bandwidth resource allocation decision and the computing resource allocation decision, calculate the average satisfaction degree; Select the offloading decision with the highest average satisfaction and name it the offloading decision , update the network parameters of the task offloading network according to the following formula :[[]]END]] ； In the above formula, represents the learning rate of the task offloading network; represents the gradient with respect to the task offloading network parameter ; S3. Compute offloading decision The satisfaction of a single task in and name it as the reward. Compute the satisfaction of a single task when all tasks are executed locally and name it as the reward and obtain the offloading decision according to the following formula The offloading reward of a single task in : ； S4. Combine to obtain a quadruple , where is the input data for time slot . Input the quadruple into the Critic network, and the Critic network outputs a vector with a size of , indicating the number of edge server caching services. The elements in the vector are , indicating the value prediction value of selecting service for the input data . S5. Use to update the network parameters of the Critic network and the network parameters of the Actor network The update formula is: ； ； ； In the above formula, represents the learning rate of the Critic network; represents the gradient with respect to the Critic network parameter ; represents the probability distribution of selecting an action according to the Actor network parameter under the given ; represents the target value of the input data ; represents the discount factor.

6. According to a railway edge computing service cache and task offloading system as described in claim 5, wherein: The objective function of the railway system optimization model is: ； ； ； ； ； ； ； ； In the above formula, represents the average satisfaction level; represents the computing resource vector; represents the service cache vector; represents the bandwidth allocation vector, represents the task offloading vector; represents the task 's satisfaction level; represents the scheduling period; represents the number of user devices; represents the task 's maximum tolerable delay; represents the task 's offloading decision, represents the task executed locally, represents the task offloaded to the edge server for execution; represents the computing resources allocated to the task ; represents the bandwidth resources allocated to the task in a single time slot; represents the satisfaction correction parameter; represents the total completion time of the offloaded tasks; represents the task 's total processing time when executed locally; represents the service 's caching decision. When represents the service cached at the edge, and when represents the service not cached; represents the wireless communication transmission delay of the task from the local device to the edge server; represents the service 's download time from the cloud to the edge server; represents the task 's execution delay; 、 respectively represent the caching status of the service required by the task in time slot 、time slot ; represents the transmission rate between the edge server and the cloud; represents the uplink data rate; represents the wireless channel gain between the local device and the edge server for the task in time slot ; represents the transmit power of the local device; represents the noise power spectral density; represents the task input data size; represents the service storage requirement; represents the computing resources required for the task; represents the task in the time slot requests service indicator variable; represents the average processing frequency of the user equipment; The constraint conditions of the railway system optimization model include: ； ； ； ； ； ； ； ； ； ； ； ； In the above formula, represents the total energy consumption of the edge server in time slot ; represents the energy consumption of task executed on the edge server in time slot ; represents the maximum allowable energy consumption of the edge server; represents the energy consumption of task transmitted on the edge server in time slot ; represents the total computing power of the edge server; represents the maximum cache capacity of the edge server; represents the number of services cached in the edge server; represents the total bandwidth of the edge server; represents the average static power consumption of service ; represents the effective switching capacitance coefficient related to the edge server.

7. According to a railway edge computing service cache and task offloading system as described in claim 6, wherein: The objective function of the bandwidth resource allocation model is: ； ； In the above formula, represents the task in the satisfaction after deconstruction; The constraint condition of the bandwidth resource allocation model is: ； ； ； ； ； ； ； In the above formula, represents the Lambert W function; represents the computing task the minimum bandwidth resource required under the condition of meeting its maximum tolerable delay; represents the maximum bandwidth resource required for the task when the service is not cached and needs to be downloaded from the cloud; and respectively represent the maximum and minimum bandwidth resources allocated to the task in a single time slot.

8. According to a railway edge computing service cache and task offloading system as described in claim 6 or 7, wherein: The objective function of the computing resource allocation model includes: ； ； The constraint conditions of the computing resource allocation model include: Minimum resource constraint: ； In the above formula, represents the task in the time slot required minimum computing resources; Energy threshold constraint: ； Computing resource upper limit constraint: ； Introduce the Lagrangian function into the computing resource allocation model to convert the computing resource allocation model into a convex problem for solution: ； In the above formula, is the Lagrangian function; , , respectively represent the Lagrange multipliers related to the minimum resource constraint, the energy threshold constraint, and the upper limit constraint of computing resources.

Images