Information processing method and device based on rail transit, and electronic equipment

Abstract

The invention discloses an information processing method and device based on rail transit and electronic equipment, and the method comprises the steps: obtaining a local execution function and an edge unloading function, and constructing a cost function of task unloading through the local execution function and the edge unloading function; the cost function is input into a deep neural network model, a target unloading scheme is output, and the task execution cost of the target unloading scheme is the lowest execution cost; a plurality of parallel deep neural networks are applied, a plurality of unloading schemes of a calculation task are stored in a shared memory and introduced into an advanced deep learning model, and an optimal unloading scheme can be finally determined through a continuous iteration process; compared with a traditional single deep neural network, an optimal edge unloading scheme can be solved more quickly and is closer to a target function, so that the convergence of edge unloading is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to an information processing method, apparatus, electronic device, and storage medium based on rail transit. Background Technology

[0002] my country's urban rail transit has developed rapidly, and train communication networks have evolved from a single monitoring function to a system integrating monitoring, diagnosis, control, and multi-disciplinary support. Train formation has also shifted from fixed to dynamic multiple-unit and flexible formations. The "Outline for the Development of Smart Urban Rail Transit in China," released by the China Urban Rail Transit Association, proposes the concept of "smart trains," emphasizing the integration of multi-system control and the sharing of multi-resource information. The TCMS (Train Network Control System) needs to integrate operations control, video surveillance, multimedia playback, and intelligent operation and maintenance to provide more convenient, safe, and efficient passenger services.

[0003] Mobile edge computing research mainly falls into two categories: offloading methods that reduce latency and offloading methods that reduce energy consumption. For example, the IHRA and LODCO algorithms reduced offloading task execution time by 44% and 64%, respectively. However, these methods only focus on latency and do not consider the energy consumption of mobile terminals, which may lead to insufficient power affecting offloading strategies and impacting user experience. Therefore, it is necessary to explore offloading decision-making schemes that minimize energy consumption, such as the artificial fish swarm algorithm to optimize offloading strategies and consider network link conditions to reduce energy consumption, although this suffers from the problem of excessive algorithm complexity. These methods only reduce energy consumption but do not achieve the goal of minimizing task execution latency. Summary of the Invention

[0004] The purpose of this invention is to provide an information processing method, device, electronic device, and storage medium based on rail transit, so as to solve the technical problem of poor adaptability of existing information processing based on rail transit.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an information processing method based on rail transit, comprising:

[0006] Obtain the local execution function and the edge unloading function, and use the local execution function and the edge unloading function to construct the cost function for task unloading;

[0007] The cost function is input into a deep neural network model, which outputs a target unloading scheme. The task execution cost of the target unloading scheme is the minimum execution cost.

[0008] In one possible implementation, the process of obtaining the local execution function is as follows:

[0009] Obtain the first energy consumption and first delay parameters for the locally executed subtask;

[0010] A local execution function is established using the first energy consumption and the first delay parameter;

[0011] The formula for the local execution function is as follows: ;

[0012] in, Here, γ1 and γ2 are weighting coefficients, where γ1 is linearly related to the maximum execution time Tmax in the locally executed subtask, and γ2 is linearly related to the maximum energy consumption Emax during local execution. For the first delay parameter, It is the primary energy consumer.

[0013] In one possible implementation, obtaining the first energy consumption and first delay parameters of the locally executed subtask includes:

[0014] Get the execution amount of the subtask, the energy consumption per byte of executing one byte locally, and the number of clock cycles per byte of executing one byte locally;

[0015] The task resource consumption of the subtask is obtained based on the amount of tasks executed and the number of clock cycles per byte.

[0016] Based on the task resource consumption and energy consumption per byte of the subtask, the first energy consumption of the subtask is determined;

[0017] Obtain the operating frequency of the device;

[0018] Based on the operating frequency and the task resource usage of the subtask, the first delay parameter of the subtask is determined.

[0019] In one possible implementation, the process of obtaining the edge unloading function is as follows:

[0020] Obtain the second energy consumption and second delay parameters of the edge unloading subtask;

[0021] An edge unloading function is established using the second energy consumption and the second delay parameter;

[0022] The formula for the edge unloading function is as follows: ;

[0023] in, The edge unloading function is defined, with γ3 and γ4 as weighting parameters. γ3 is linearly related to the maximum execution time TD of the subtask in the Manufacturing Execution System (MES), and γ4 is linearly related to the maximum energy consumption EM during the edge unloading process. For the second delay parameter, This is the second largest energy consumption.

[0024] In one possible implementation, obtaining the second energy consumption and second delay parameters of the edge offloading subtask includes:

[0025] Obtain the third energy consumption of MES task execution and the fourth energy consumption generated during task compilation;

[0026] Calculate the sum of the third and fourth energy consumptions to obtain the second energy consumption;

[0027] Obtain the uplink time, execution time, downlink time, and compilation time during the process of unloading subtasks to MES;

[0028] The second delay parameter is obtained by summing the uplink time, execution time, downlink time, and compilation time.

[0029] In one possible implementation, the uplink and downlink times during the subtask unloading process into MES are obtained as follows:

[0030] Orthogonal frequency division multiple access (OFDM) technology is used to complete the allocation of communication resources, and the noise performance of the signal processing system after allocation is analyzed to obtain the maximum data transmission rate of the uplink and downlink.

[0031] By dividing the amount of work performed by the subtask by the maximum data transmission rates of the uplink and downlink respectively, the uplink and downlink times during the process of offloading the subtask to the MES are obtained.

[0032] In one possible implementation, the deep neural network model comprises multiple parallel deep neural networks, such that the deep neural network model iterates repeatedly to output a target unloading scheme with the lowest task execution cost.

[0033] Secondly, the present invention provides an information processing device based on rail transit, comprising:

[0034] The acquisition module is used to acquire the local execution function and the edge unloading function, and to construct the cost function for task unloading using the local execution function and the edge unloading function;

[0035] An execution module is used to input the cost function into a deep neural network model and output a target unloading scheme, wherein the task execution cost of the target unloading scheme is the lowest execution cost. In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described information processing method based on rail transit.

[0036] Fourthly, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described information processing method based on rail transit.

[0037] The technical effects and advantages of this invention are as follows:

[0038] The technical solution of this application establishes a novel deep learning model. By using multiple parallel deep neural networks and storing multiple offloading schemes of the computational task in a shared memory, it introduces an advanced deep learning model. Through continuous iteration, the optimal offloading scheme can be determined. Compared with the traditional single deep neural network, it can solve the optimal edge offloading scheme more quickly and is closer to the objective function, thereby significantly improving the convergence of edge offloading. Attached Figure Description

[0039] Figure 1 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 1 ;

[0040] Figure 2 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 2 ;

[0041] Figure 3 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 3 ;

[0042] Figure 4 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 4 ;

[0043] Figure 5 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 5 ;

[0044] Figure 6 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 6 ;

[0045] Figure 7 An edge computing architecture diagram of a deep learning model provided in an embodiment of this application;

[0046] Figure 8 A schematic diagram illustrating the learning process of a DNN provided in an embodiment of this application;

[0047] Figure 9 An edge computing model diagram of a train wireless device network provided in this application embodiment;

[0048] Figure 10 This is a schematic diagram illustrating the effect of the bandwidth quantity provided in the embodiments of this application on the transmission delay and calculation delay during offloading;

[0049] Figure 11 This is a schematic diagram illustrating the effect of the number of CPUs on transmission and computation delays during unloading, provided in an embodiment of this application.

[0050] Figure 12 A schematic diagram illustrating the communication delay for offloading different algorithms provided in embodiments of this application;

[0051] Figure 13 A schematic diagram illustrating the computation delay of MES using different algorithms provided in the embodiments of this application;

[0052] Figure 14 This is a schematic diagram of the structure of an information processing device based on rail transit provided in an embodiment of this application;

[0053] Figure 15 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0055] The following disclosure provides numerous different embodiments or examples for implementing various structures of this embodiment. To simplify the disclosure of this embodiment, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the scope of this embodiment. Furthermore, reference numerals and / or letters may be repeated in different examples of this embodiment. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0056] This application presents an energy- and time-sensitive deep learning-based unloading algorithm. This algorithm consists of two parts: developing an unloading scheme and deep learning, thus solving the problem of selective unloading of mobile application components. The innovation of this algorithm is mainly reflected in the following two points:

[0057] First, when establishing the system utility model of the MEC network, the main operators are delayed communication and remaining energy, which are weighted to derive the offloading cost. Based on this, to improve offloading efficiency, this embodiment also fully considers factors closely related to offloading efficiency, such as computing resources and communication resources. By optimizing the computing load and communication resources allocated to each computing task, the most reasonable cost function is constructed during the edge offloading process, thus obtaining the corresponding offloading scheme.

[0058] Secondly, a novel deep learning model is established. This model uses multiple parallel deep neural networks to store several offloading schemes for the computational task in a shared memory. These schemes are then substituted into the novel deep learning model and iterated repeatedly to finally obtain the optimal offloading scheme. Compared with the traditional single deep neural network, the advantage is that it can solve the optimal edge offloading scheme more quickly. The embodiments of this application are closer to the objective function, that is, the convergence of edge offloading is greatly improved.

[0059] Figure 1 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 1 ,like Figure 1 As shown, the method specifically includes:

[0060] S101. Obtain the local execution function and the edge unloading function, and use the local execution function and the edge unloading function to construct the cost function for task unloading;

[0061] In this embodiment, by obtaining the local execution function and the edge unloading function, a detailed analysis of the cost function of task unloading is further used to construct a task unloading process, ensuring that the local execution function and the edge unloading function can work together to optimize resource allocation and performance during the task unloading process.

[0062] S102. Input the cost function into the deep neural network model and output the target unloading scheme, wherein the task execution cost of the target unloading scheme is the minimum execution cost.

[0063] In this embodiment, by using a specific cost function as input data, the deep neural network model can process and analyze this information, and then output an optimized target unloading scheme. This target unloading scheme can achieve the lowest execution cost when performing the task, ensuring efficient use of resources and minimization of costs.

[0064] The information processing method for rail transit provided in this application combines local execution functions and edge unloading functions, and utilizes a deep neural network model to achieve intelligent optimization of task unloading schemes. It not only considers the energy consumption and latency of task execution, but also gradually approximates the optimal unloading strategy through an iterative process of deep learning, thereby ensuring the efficiency and cost-effectiveness of task execution. Furthermore, the innovation of this method lies in its use of multiple parallel deep neural networks, which accelerates the solution process for the optimal unloading scheme through parallel processing, further improving the convergence and practicality of edge unloading.

[0065] Figure 2 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 2 ,like Figure 2As shown, the process of obtaining the local execution function is as follows:

[0066] S201. Obtain the first energy consumption and first delay parameters of the locally executed subtask;

[0067] S202. Using the first energy consumption and the first delay parameter, establish a local execution function;

[0068] In this embodiment, firstly, the first energy consumption parameters generated during the local execution of the subtask are obtained, such as detailed data on processor utilization and other related energy consumption during task execution. Simultaneously, the time required for the task to complete from start to finish is recorded, including any possible delays and interruptions. Then, using these first energy consumption and first delay parameters, a local execution function is established through in-depth analysis and calculation.

[0069] The formula for the local execution function is as follows: ;

[0070] in, Here, γ1 and γ2 are weighting coefficients, where γ1 is linearly related to the maximum execution time Tmax in the locally executed subtask, and γ2 is linearly related to the maximum energy consumption Emax during local execution. For the first delay parameter, It is the primary energy consumer.

[0071] For example, the locally executed function, denoted as C, is derived from the computational latency and energy consumption of locally executed functions. l (s):

[0072] ;

[0073] In the above formula, the weighting coefficients γ1 and γ2 are linearly related to the maximum execution time Tmax in the subtask and the maximum energy consumption Emax in the task, respectively.

[0074] Furthermore, to address issues such as limited computing resources and battery power, computationally intensive and latency-sensitive tasks on mobile devices can be offloaded from the local machine to edge servers for execution, thereby improving system performance. Similar to local execution, computation latency and energy consumption remain the two most important operators affecting edge offloading. These two operators are denoted as Tr(c) and [the following operators are also mentioned:] Taking full account of the impact of computing and communication resources on Tr(c) and Based on the resulting impact, the most reasonable cost function is formulated to obtain the optimal unloading scheme.

[0075] The information processing method based on rail transit provided in this application establishes corresponding local execution functions and edge unloading functions by deeply analyzing the energy consumption and latency parameters of local execution and edge unloading. Based on this, a cost function for task unloading is further constructed using these functions, taking into account various resource consumption and performance during task execution to ensure the efficiency and cost-effectiveness of task execution.

[0076] Figure 3 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 3 ,like Figure 3 As shown,

[0077] The process of obtaining the first energy consumption and first latency parameters of the locally executed subtask includes:

[0078] S301, obtain the execution task quantity of the subtask, the energy consumption per byte of executing one byte locally, and the number of clock cycles per byte occupied by executing one byte locally;

[0079] S302. Obtain the task resource usage of the subtask based on the amount of task executed and the number of clock cycles per byte.

[0080] S303. Based on the task resource consumption and energy consumption per byte of the subtask, determine the first energy consumption of the subtask;

[0081] S304. Obtain the operating frequency of the device.

[0082] S305. Based on the operating frequency and the task resource usage of the subtask, determine the first delay parameter of the subtask.

[0083] In this embodiment, the cost function of any subtask t in any WD in the model, which is executed locally, is first determined by the computation delay and energy consumption, and then derived using these two operators;

[0084] Assumption: The data size of any subtask of WD is d nt W t Let V represent the computing resources used by subtask t, V represent the number of clock cycles required for the CPU to execute one byte, and fu represent the CPU operating frequency of the wireless device; therefore, the computation latency is:

[0085] ;

[0086] Among them W t The data size d of the subtask nt We obtain the following from the number of clock cycles V required for the CPU to execute one byte:

[0087] ;

[0088] During local execution, the energy consumed in executing each byte is denoted as . The energy consumed in executing the subtask is calculated as follows:

[0089] ;

[0090] The information processing method based on rail transit provided in this application offers detailed parameter acquisition and calculation methods. First, the execution workload of the subtask is obtained, which is the basis for calculating resource consumption and energy usage. Then, by combining the energy consumption per byte of local execution and the number of clock cycles per byte, the energy consumption of the subtask during local execution can be accurately calculated. Furthermore, by obtaining the operating frequency of the device and combining it with the resource consumption of the subtask, the first delay parameter of the subtask can be reasonably estimated, providing strong support for constructing a reasonable cost function and thus ensuring the optimization effect of the task offloading scheme.

[0091] Figure 4 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 4 ,like Figure 4 As shown, the process of obtaining the edge unloading function is as follows:

[0092] S401. Obtain the second energy consumption and second delay parameters of the edge unloading subtask;

[0093] S402. Using the second energy consumption and the second delay parameter, establish an edge unloading function;

[0094] The formula for the edge unloading function is as follows:

[0095] ;

[0096] in, The edge unloading function is defined, with γ3 and γ4 as weighting parameters. γ3 is linearly related to the maximum execution time TD of the subtask in the Manufacturing Execution System (MES), and γ4 is linearly related to the maximum energy consumption EM during the edge unloading process. For the second delay parameter, This is the second largest energy consumption.

[0097] Figure 5 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 5 ,like Figure 5 As shown, the acquisition of the second energy consumption and second delay parameters of the edge unloading subtask includes:

[0098] S501, obtain the third energy consumption of MES executing tasks and the fourth energy consumption generated during task compilation;

[0099] S502. Calculate the sum of the third energy consumption and the fourth energy consumption to obtain the second energy consumption;

[0100] S503, Obtain the uplink time, execution time, downlink time, and compilation time during the process of unloading the subtask to MES;

[0101] S504. Calculate the sum of the uplink time, execution time, downlink time, and compilation time to obtain the second delay parameter.

[0102] In this embodiment, the process of offloading subtasks from the WD to the MES is actually as follows: first, the data is first uploaded to the MES; second, the task is executed in the MES; third, after the task is completed, the data is downloaded back to the WD; and finally, the completed task is decoded in the WD. Therefore, four time constants are generated: the upload time of the subtask, i.e., the uplink time T. up The execution time T of subtasks in MES ex The downlink time from the subtask to WD, i.e., the downlink time T. down Decoding time T d .

[0103] ；

[0104] As can be seen from the above formula, the allocation of communication resources, i.e., bandwidth, directly affects the uplink time T of the subtask. up and downlink time T down The allocation of computing resources, i.e., CPU, directly affects the execution time T of subtasks in MES. ex The calculation formulas are as follows:

[0105] , , ;

[0106] In the above formula, r ul r dl These represent the uplink speed to MES and the downlink speed to WD for the subtask, respectively. These two factors are closely related to the allocation of communication resources; Wt represents the number of clock cycles required to execute subtask t, m represents the number of CPUs allocated to the subtask (the number of m depends on the data structure of the subtask), and f s This represents the CPU operating frequency of the MES. Bandwidth allocation directly affects the uplink time T of subtasks. up and downlink time T down .

[0107] Furthermore, in this embodiment, Orthogonal Frequency Division Multiple Access (OFDMA) technology is used to allocate communication resources. The total network bandwidth B is decomposed into K subcarriers (communication resources, channels), and each subtask t is allocated to several subcarriers (the number is K, k∈1,2,3,...,K). Since additive white Gaussian noise (AWGN) is easy to analyze and approximate, in the field of signal processing, in simple analysis of the noise performance of signal processing systems, the actual noise signal is often approximated only within a certain frequency band using the characteristics of white Gaussian noise. Therefore, the maximum data transmission rates for the uplink and downlink can be derived as follows:

[0108] ;

[0109] ;

[0110] In the above formula, n is the carrier parameter, N is the noise parameter, B is the network bandwidth, D is the distance from WD to MES, N0 is the noise power consumption, and P... u h represents the power consumption generated by WD transmitting data. ul β is the attenuation coefficient of the uplink channel, β is the path loss exponent, and g is the path loss coefficient. ul These are errors in uplink data transmission. Γ represents the stability margin of the signal-to-noise ratio (SNR), which aims to meet the target bit error rate. Γ is the adjustment method for calculating the SNR, and gul is the uplink interference parameter.

[0111] Similar to the analysis method for calculating latency, the energy consumption generated by offloading subtasks to MES is also considered. The energy consumption E mainly comes from the tasks performed by the MES. ex Energy consumption E for decoding in WD d The energy consumption generated by data uplink and downlink is negligible.

[0112] ;

[0113] The information processing method for rail transit provided in this application establishes a corresponding function model by deeply analyzing the energy consumption and delay parameters of local execution and edge unloading. Based on this model, a cost function for task unloading is constructed. This method not only considers various resource consumptions and performance characteristics during task execution but also gradually approximates the optimal unloading strategy through an iterative process of deep learning. Furthermore, the method employs multiple parallel deep neural networks for parallel processing, thereby accelerating the solution process for the optimal unloading scheme and further improving the convergence and practicality of edge unloading.

[0114] Figure 6 The flowchart of the information processing method based on rail transit provided in the embodiments of this application Figure 6 ,like Figure 6 As shown, the methods for obtaining the uplink and downlink times during the subtask unloading process to MES are as follows:

[0115] S601. Using orthogonal frequency division multiple access (OFDMA) technology, communication resources are allocated, and the noise performance of the signal processing system after allocation is analyzed to obtain the maximum data transmission rate of the uplink and downlink.

[0116] S602. Divide the execution amount of the subtask by the maximum data transmission rate of the uplink and downlink respectively to obtain the uplink time and downlink time during the process of offloading the subtask to MES.

[0117] In this embodiment, the allocation of communication resources is effectively completed by applying Orthogonal Frequency Division Multiple Access (OFDMA) technology. Based on this, the noise performance of the allocated signal processing system is further analyzed in depth to accurately calculate the maximum data transmission rate that the uplink and downlink can support respectively.

[0118] After determining the maximum data transmission rates for the uplink and downlink, the uplink and downlink times of the subtasks during the offloading process to the Manufacturing Execution System (MES) are accurately estimated by dividing the execution workload of each subtask by the calculated maximum data transmission rates, thereby providing data support for the optimization of the entire communication system.

[0119] Furthermore, mobile devices require a certain amount of latency and energy to perform corresponding computing tasks. To minimize latency and energy consumption and facilitate analysis, a system utility model S is introduced. This model includes four key parameters that affect system utility, expressed as follows:

[0120] {S=(d,k,m,E nt) ∈S|d∈D,k∈K,m∈M};

[0121] In the above formula, S represents the state model of system utility. This vector contains four key parameters for performing the computational task: the number of bytes d in the subtask, the allocated communication resources, k subcarriers, the allocated computing resources, m CPUs, and the energy consumption E for performing the task. nt .

[0122] Based on the above cost analysis of local and edge unloading, and the established system utility model, the cost function for performing computational tasks is finally derived as follows:

[0123] ;

[0124] In the above formula, P t， Represents the binary offloading scheme for each subtask in WD; P t =0 indicates that the WD subtask is executed locally, P t =1 indicates that WD's subtasks should be offloaded to the edge server for execution.

[0125] Based on the system model S and cost function for determining the task to be performed. Based on this, we find the optimal unloading scheme Po that minimizes the cost of executing tasks, and obtain the most reasonable resource allocation scheme, which is the optimal solution for mobile edge computing.

[0126] In the process of seeking the optimal solution, a Deep Neural Network (DNN) is used. The system utility parameters of the computational task are input into the DNN, and after corresponding matrix operations and discrimination in the hidden layers, the unloading scheme is obtained in the output layer. The unloading scheme is then subjected to backpropagation and iteration (continuously performing the backward gradient reduction method) of the neural network to find the optimal solution Po that is closest to the actual value. Its mathematical expression is as follows:

[0127] ;

[0128] In the above formula, Po represents the total cost of performing the computational task, P represents the true optimal unloading scheme, and the closest value to the optimal solution obtained by using a deep neural network (DNN) is to make Po infinitely close to P, that is: the arctangent value arg approaches 0.

[0129] The model design is as follows:

[0130] like Figure 7 As shown, Deep Neural Networks (DNNs) are the foundation of deep learning. DNNs achieve high accuracy when training on large datasets. In this embodiment, multiple parallel deep neural networks (DNNSs) are used to train the samples and derive the optimal unloading scheme. The input layer of this model is a set S* of vector states of several subtasks in the computation task, i.e.:

[0131] ;

[0132] The output layer is an offloading scheme P. In this embodiment, the computational task of WD is divided into several subtasks, so P = Pt, c = 1, 2, 3... |k|. Each subtask has two possible offloading scenarios: execution locally (Pt = 0) or execution by MES (Pt = 1). Therefore, there are t subtasks. One uninstallation solution.

[0133] Figure 8 This is a schematic diagram of the learning process of a DNN provided in an embodiment of this application. Figure 8 In this context, w represents the weights of the hidden layer, and σ() is the activation function. , The cross-entropy loss function is used. The learning process consists of two steps: forward propagation and backward propagation.

[0134] The information processing method based on rail transit provided in this application uses a parallel deep neural network (DNN). This method increases the number of hidden layers and neurons in the network, setting the number of hidden layers to two and the number of neurons in each layer to 256. Numerical analysis in this embodiment shows that the more hidden layers and neurons there are, the higher the computational accuracy. This embodiment uses the sigmoid function as the activation function, with a value range of [0,1], for the output of the hidden layer neurons. Finally, the cross-entropy loss function is used as the loss function. Its advantage is that using the sigmoid function avoids the mean squared error loss during gradient descent, which can lead to a decrease in the learning rate. Therefore, the learning rate can be controlled by the output error.

[0135] like Figures 9 to 13 As shown, to better evaluate the performance of the EDLO algorithm, its performance is compared with that of four other computational task processing schemes. In this embodiment, the EDLO algorithm uses 10 DNNS, a learning rate of 0.01, and 2048 MB of system memory to achieve optimal performance. The other four schemes are as follows:

[0136] Total local process (TLP): All computing tasks of the wireless device (WD) are executed locally.

[0137] Total Edge Processing (TEP): All computing tasks of the wireless device (WD). [31-32] This is executed by the edge server MES;

[0138] Random offloading scheme (ROS): Without considering the state space S of the computation task, subtasks are randomly assigned to edge servers or local wireless devices for processing.

[0139] The Deep Learning Based Offloading Scheme (DLO) is an algorithm whose input states do not include the remaining energy of the wireless device or the energy consumption of the execution program. The neural network of this algorithm contains two layers and two hidden layers, each with 128 neurons.

[0140] To better verify the performance of the EDLO algorithm, with the same network parameters, this embodiment conducted the following comparative tests on the EDLO algorithm and four other algorithms:

[0141] The impact of communication resources on offloading delay, i.e., the effect of bandwidth B on T. ud =T up +T down The impact;

[0142] The impact of computing resources on edge computing latency, i.e., the effect of the number of CPUs m on T. ex The impact;

[0143] The total computational latency generated during task execution is compared, i.e., the total latency consumption of task execution;

[0144] Accuracy comparison;

[0145] A comparison of energy consumption generated;

[0146] Cost comparison. This comparison allows for a more comprehensive verification of the EDLO algorithm's performance.

[0147] Because of the Total Local Processing (TLP) scheme, data is processed locally without using the computing and communication resources of the MES. In this embodiment, with a network bandwidth of 100Mbps, 10 cores in the MES, and a computing task size of 15MB, the communication latency T for offloading is compared with that of four other different algorithms. ud The computational delay T of MES ex The impact;

[0148] The latency of the EDLO algorithm is significantly lower than that of the TLP and ROS algorithms, and also lower than that of the traditional deep learning algorithm DLO. The communication latency of EDLO is 30.76% of that of DLO, and the computation latency is 31.58% of that of DLO. This demonstrates that the computational efficiency of the new algorithm is significantly improved because this embodiment fully considers the impact of bandwidth and CPU on latency during modeling.

[0149] Experimental results demonstrate that this application proposes an energy- and time-sensitive deep learning offloading algorithm, EDLO. Taking into full account the system's energy consumption, computational latency, computational resources, and communication resources, this algorithm introduces multiple parallel deep neural networks. It can generate optimal solutions without manually labeled data, and numerical results verify the algorithm's accuracy and performance advantages. This embodiment anticipates that this energy- and time-sensitive deep learning offloading framework can be further extended in future MEC network implementations to optimize real-time offloading, thereby improving the network communication efficiency of trains in rail transit.

[0150] Figure 14 This is a schematic diagram of the structure of the information processing device based on rail transit provided in the embodiments of this application, as shown below. Figure 14 As shown, the device specifically includes:

[0151] The acquisition module 1401 is used to acquire the local execution function and the edge unloading function, and to construct the cost function for task unloading using the local execution function and the edge unloading function;

[0152] The execution module 1402 is used to input the cost function into the deep neural network model and output the target unloading scheme, wherein the task execution cost of the target unloading scheme is the minimum execution cost.

[0153] In one possible implementation, the acquisition module 1401 is further configured to acquire a first energy consumption and a first delay parameter of the locally executed subtask; and to establish a local execution function using the first energy consumption and the first delay parameter; wherein the formula of the local execution function is as follows: ;in, Here, γ1 and γ2 are weighting coefficients, where γ1 is linearly related to the maximum execution time Tmax in the locally executed subtask, and γ2 is linearly related to the maximum energy consumption Emax during local execution. For the first delay parameter, It is the primary energy consumer.

[0154] In one possible implementation, the acquisition module 1401 is further configured to acquire the execution task amount of the subtask, the energy consumption per byte of local execution of one byte, and the number of clock cycles per byte occupied by local execution of one byte; obtain the task resource occupancy of the subtask based on the execution task amount and the number of clock cycles per byte; determine the first energy consumption of the subtask based on the task resource occupancy and the energy consumption per byte; acquire the operating frequency of the device; and determine the first delay parameter of the subtask based on the operating frequency and the task resource occupancy of the subtask.

[0155] In one possible implementation, the acquisition module 1401 is further configured to acquire the second energy consumption and second delay parameters of the edge unloading subtask; and to establish an edge unloading function using the second energy consumption and second delay parameters.

[0156] In one possible implementation, the acquisition module 1401 is further configured to acquire the third energy consumption of the MES execution task and the fourth energy consumption generated by task compilation; calculate the sum of the third energy consumption and the fourth energy consumption to obtain the second energy consumption; acquire the uplink time, execution time, downlink time and compilation time during the process of the subtask being unloaded to the MES; calculate the sum of the uplink time, execution time, downlink time and compilation time to obtain the second delay parameter.

[0157] In one possible implementation, the acquisition module 1401 is further configured to use orthogonal frequency division multiple access (OFDM) technology to complete the allocation of communication resources, and analyze the noise performance of the signal processing system after allocation to obtain the maximum data transmission rate of the uplink and downlink; and to use the execution workload of the subtask to divide the maximum data transmission rate of the uplink and downlink respectively to obtain the uplink time and downlink time during the process of the subtask being unloaded to the MES.

[0158] In one possible implementation, the execution module 1402 is further configured to include multiple parallel deep neural networks in the deep neural network model, so that the deep neural network model iterates repeatedly to output a target unloading scheme with the lowest task execution cost.

[0159] The information processing device based on rail transit provided in this embodiment can be as follows: Figure 14 The information processing device based on rail transit shown can perform tasks such as... Figures 1-13 All steps of information processing based on rail transit in China, thereby achieving Figures 1-13 The technical effects of information processing based on rail transit shown are illustrated in the image. Please refer to [link / reference needed]. Figures 1-13 The relevant descriptions are presented concisely and will not be elaborated upon here.

[0160] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] Figure 15 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 15As shown, this application embodiment provides an electronic device, including a processor 1501, a communication interface 1502, a memory 1503, and a communication bus 1504. The processor 1501, communication interface 1502, and memory 1503 communicate with each other via the communication bus 1504. The memory 1503 stores computer programs. When the processor 1501 executes the program stored in the memory 1503, it implements the steps of the information processing method based on rail transit provided in any of the aforementioned method embodiments.

[0162] Obtain the local execution function and the edge unloading function, and use the local execution function and the edge unloading function to construct the task unloading cost function; input the cost function into the deep neural network model, and output the target unloading scheme, wherein the task execution cost of the target unloading scheme is the lowest execution cost.

[0163] In one possible implementation, a first energy consumption and a first delay parameter are obtained for the locally executed subtask; a locally executed function is established using the first energy consumption and the first delay parameter; wherein the formula of the locally executed function is as follows: ;in, Here, γ1 and γ2 are weighting coefficients, where γ1 is linearly related to the maximum execution time Tmax in the locally executed subtask, and γ2 is linearly related to the maximum energy consumption Emax during local execution. For the first delay parameter, It is the primary energy consumer.

[0164] In one possible implementation, the following steps are taken: obtaining the execution task amount of the subtask, the energy consumption per byte of local execution of one byte, and the number of clock cycles per byte occupied by local execution of one byte; obtaining the task resource occupancy of the subtask based on the execution task amount and the number of clock cycles per byte; determining the first energy consumption of the subtask based on the task resource occupancy and the energy consumption per byte; obtaining the operating frequency of the device; and determining the first delay parameter of the subtask based on the operating frequency and the task resource occupancy of the subtask.

[0165] In one possible implementation, a second energy consumption and a second delay parameter of the edge unloading subtask are obtained; using the second energy consumption and the second delay parameter, an edge unloading function is established; wherein the formula of the edge unloading function is as follows: ;in, The edge unloading function is defined, with γ3 and γ4 as weighting parameters. γ3 is linearly related to the maximum execution time TD of the subtask in the Manufacturing Execution System (MES), and γ4 is linearly related to the maximum energy consumption EM during the edge unloading process. For the second delay parameter, This is the second largest energy consumption.

[0166] In one possible implementation, the third energy consumption of the MES execution task and the fourth energy consumption generated by task compilation are obtained; the sum of the third energy consumption and the fourth energy consumption is calculated to obtain the second energy consumption; the uplink time, execution time, downlink time and compilation time of the subtask being unloaded to the MES are obtained; the sum of the uplink time, execution time, downlink time and compilation time is calculated to obtain the second delay parameter.

[0167] In one possible implementation, orthogonal frequency division multiple access (OFDM) technology is used to complete the allocation of communication resources, and the noise performance of the signal processing system after allocation is analyzed to obtain the maximum data transmission rates of the uplink and downlink. The uplink and downlink times during the subtask offloading to the MES are obtained by dividing the execution workload of the subtask by the maximum data transmission rates of the uplink and downlink, respectively.

[0168] In one possible implementation, the deep neural network model comprises multiple parallel deep neural networks, such that the deep neural network model iterates repeatedly to output a target unloading scheme with the lowest task execution cost.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0170] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0171] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in these embodiments may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown in this embodiment, but is to be accorded the widest scope consistent with the principles and novel features claimed in this embodiment.

Claims

1. An information processing method based on rail transit, characterized in that, include: Obtain the local execution function and the edge unloading function, and use the local execution function and the edge unloading function to construct the cost function for task unloading; The cost function is input into a deep neural network model, which outputs a target unloading scheme. The task execution cost of the target unloading scheme is the minimum execution cost.

2. The method according to claim 1, characterized in that, The process of obtaining the local execution function is as follows: Obtain the first energy consumption and first delay parameters for the locally executed subtask; A local execution function is established using the first energy consumption and the first delay parameter; The formula for the local execution function is as follows: ； in, Here, γ1 and γ2 are weighting coefficients, where γ1 is linearly related to the maximum execution time Tmax in the locally executed subtask, and γ2 is linearly related to the maximum energy consumption Emax during local execution. For the first delay parameter, It is the primary energy consumer.

3. The method according to claim 2, characterized in that, The process of obtaining the first energy consumption and first latency parameters of the locally executed subtask includes: Get the execution amount of the subtask, the energy consumption per byte of executing one byte locally, and the number of clock cycles per byte of executing one byte locally; The task resource consumption of the subtask is obtained based on the amount of tasks executed and the number of clock cycles per byte. Based on the task resource consumption and energy consumption per byte of the subtask, the first energy consumption of the subtask is determined; Obtain the operating frequency of the device; Based on the operating frequency and the task resource usage of the subtask, the first delay parameter of the subtask is determined.

4. The method according to claim 1, characterized in that, The process of obtaining the edge unloading function is as follows: Obtain the second energy consumption and second delay parameters of the edge unloading subtask; An edge unloading function is established using the second energy consumption and the second delay parameter; The formula for the edge unloading function is as follows: ； in, The edge unloading function is defined, with γ3 and γ4 as weighting parameters. γ3 is linearly related to the maximum execution time TD of the subtask in the Manufacturing Execution System (MES), and γ4 is linearly related to the maximum energy consumption EM during the edge unloading process. For the second delay parameter, This is the second largest energy consumption.

5. The method according to claim 4, characterized in that, The acquisition of the second energy consumption and second delay parameters of the edge unloading subtask includes: Obtain the third energy consumption of MES task execution and the fourth energy consumption generated during task compilation; Calculate the sum of the third and fourth energy consumptions to obtain the second energy consumption; Obtain the uplink time, execution time, downlink time, and compilation time during the process of unloading subtasks to MES; The second delay parameter is obtained by summing the uplink time, execution time, downlink time, and compilation time.

6. The method according to claim 5, characterized in that, The methods for obtaining the uplink and downlink times during the subtask unloading process to MES are as follows: Orthogonal frequency division multiple access (OFDM) technology is used to complete the allocation of communication resources, and the noise performance of the signal processing system after allocation is analyzed to obtain the maximum data transmission rate of the uplink and downlink. By dividing the amount of work performed by the subtask by the maximum data transmission rates of the uplink and downlink respectively, the uplink and downlink times during the process of offloading the subtask to the MES are obtained.

7. The method according to claim 1, characterized in that, The deep neural network model includes multiple parallel deep neural networks, which iterate repeatedly to output the target unloading scheme with the lowest task execution cost.

8. An information processing device based on rail transit, characterized in that, include: The acquisition module is used to acquire the local execution function and the edge unloading function, and to construct the cost function for task unloading using the local execution function and the edge unloading function; The execution module is used to input the cost function into the deep neural network model and output the target unloading scheme, wherein the task execution cost of the target unloading scheme is the minimum execution cost.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the information processing method based on rail transit as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the information processing method based on rail transit as described in any one of claims 1 to 7.

Images