Low-complexity calculation unloading method and system based on high-speed rail train-ground communication scene

Abstract

The invention discloses a low-complexity calculation unloading method and system based on a high-speed rail train-ground communication scene. The low-complexity calculation unloading method comprises the steps that an unloading communication integrated system architecture based on high-speed rail data tasks is constructed; determining an energy transmission protocol between each subnet AP and a sensor, and expressing a wireless energy transmission process by adopting a segmented nonlinear model; establishing a data task unloading mathematical model of each node, and calculating and acquiring time and consumed energy of task calculation of base station calculation and local calculation of the sensor through the data task unloading mathematical model of each node; constructing an overall optimization objective function, decomposing the overall optimization objective function into a power allocation optimization objective of the AP, a base station computing resource allocation optimization objective and a sensor task unloading decision optimization objective, and solving optimization variables; designing a joint optimization algorithm, and alternately optimizing and updating each optimization variable to obtain an optimal task unloading strategy; the problem that the energy of the sensors is limited is solved, the unloading effectiveness of the sensors is improved, and an efficient and reliable communication means is provided for the multiple sensors.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless energy collection and data computing task offloading and resource allocation, and more specifically to a low-complexity computing offloading method and system based on a high-speed rail vehicle-to-ground communication scenario. Background Art

[0002] As a convenient and green public transportation system, high-speed railways are developing rapidly and will become the future trend of global railway transportation in many countries. With the development of new intelligent technologies, high-speed railways will inevitably move from information-based high-speed railways to intelligent high-speed railways.

[0003] In order to ensure the reliable and safe operation of high-speed rail, intelligent high-speed rail uses Internet of Things technology to monitor different facilities and equipment in the high-speed rail system, the operating status of high-speed rail trains and the surrounding environment in real time, so as to achieve timely detection and effective handling of emergencies and achieve the purpose of improving operational safety.

[0004] However, as the number of wireless sensing devices continues to increase, the problem of continuous energy supply has become a bottleneck in the development of wireless sensing networks, limiting the large-scale deployment of sensors. At the same time, due to the limited computing resources of sensors, not only the energy limitation problem must be considered, but many sensors involve delay-sensitive tasks such as identification and calculation of train safety information. In line with this, mobile edge computing MEC has become a promising paradigm. It can reduce the load of the core network and data transmission delay by deploying nodes with computing processing capabilities at the edge of the network to be closer to users. However, in the high-speed rail scenario, there is very limited work on the combination of MEC and train-ground communication systems. In addition, the signal transmission is greatly attenuated due to the metal shielding wires in the carriage.

[0005] At present, there are few studies at home and abroad on the power supply for sensors in high-speed rail scenarios and the joint design with sensor data task offloading, and the research on sensor task offloading and resource allocation.

[0006] Therefore, how to solve the problem of limited energy and computing resources of sensor nodes and the problem of offloading strategy, and optimize the performance of wireless sensor systems, is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0007] In view of this, the present invention provides a low-complexity computing offloading method and system based on a high-speed rail vehicle-ground communication scenario to solve some of the technical problems mentioned in the background technology.

[0008] In order to achieve the above object, the present invention adopts the following technical solution:

[0009] A low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario comprises the following steps:

[0010] S1. Build an integrated system architecture based on high-speed rail data task offloading communication, including multiple base stations and carriage sub-networks. Each sub-network includes an access node AP and multiple passive sensors. In each sub-network, networking is achieved based on orthogonal frequency division multiplexing random access. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station.

[0011] S2. Determine the energy transmission protocol between the AP and the sensor within each subnet, use a piecewise nonlinear model to describe the wireless energy transmission process, obtain the receiving power of each sensor in each compartment, the subnet access node AP transmits the radio frequency signal, the sensor collects energy, and each sensor selects the calculation task to be offloaded to a base station for calculation or performs sensor local calculation;

[0012] S3. Establish a mathematical model for data task offloading at each node, including a sensor local calculation model, a car data task offloading model, a car data task offloading model, and a base station calculation offloading task model, and calculate the time and energy consumption of the sensor for base station calculation and local calculation through the mathematical model of data task offloading at each node;

[0013] S4. Construct an overall optimization objective function based on the weighted sum minimization of the time and energy consumed by the sensor task calculations in the system;

[0014] S5. Decompose the overall optimization goal into the power allocation optimization goal of each subnet access point AP, the base station computing resource allocation optimization goal and the sensor task offloading decision optimization goal, and solve the optimization variables of each optimization goal;

[0015] S6. Design a joint optimization algorithm to alternately optimize and update each optimization variable, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

[0016] Preferably, in step S2, the specific content of mathematically describing the wireless energy transmission process is:

[0017] Using the piecewise nonlinear energy harvesting model, the received power of the mth sensor in the nth compartment is calculated as:

[0018]

[0019] Among them, ε n，m ∈(0, 1] is the conversion factor, p n is the transmission power of the nth AP to the sensor in its compartment, h n，m is the channel gain between the nth AP and the mth sensor, P th is the peak power available to the mth sensor.

[0020] Preferably, the sensor local calculation model established in step S3 is specifically:

[0021] When sensor m performs a task locally, the time to complete the task is:

[0022]

[0023] Sensor m performs task T locally m The energy consumption is:

[0024]

[0025] Among them, c m is the workload, i.e. the amount of computation required to complete the task, is the local computing power of the sensor, and κ is the energy coefficient that depends on the chip architecture.

[0026] Preferably, the in-car data task unloading model established in step S3 is specifically:

[0027] The power of the sensor to transmit uploaded data is:

[0028]

[0029] Among them, the remaining energy (1-α n，m ) n，m is offset by background noise and self-interference, α n，m ∈(0, 1] is a PS constant, P n，m is the received power of the mth sensor in the nth compartment, U is the set of sensors, and N is the set of compartments;

[0030] The sensor sends data to the kth base station BS at the first hop k The achievable rate is:

[0031]

[0032] Where W is the size of dividing the operating frequency band B into M equal sub-bands, It means that sensor m is connected to the nth access point AP via sub-band j. n The first hop signal-to-noise ratio is the transmission of the computing task from the sensor m to the access point AP;

[0033]

[0034] in, is the nth access point AP on sub-band j n The channel gain between the mth sensor, G t (m, n, j), G r(m, n, j) are the distance from sensor m to the nth access point AP n The transmission gain and receiving gain of the directional antenna, βp n is the residual SI value, β is the SI cancellation level of the access point, which is a non-negative parameter, and p n The nth access point AP n The transmission power, that is, the transmission power of the wireless radio frequency radiated by the sensor in the car, σ 2 is the background noise variance;

[0035] The off-carriage data task offloading model established in step S3 is specifically:

[0036] Sensor m sends data to the kth base station BS at the second hop k The achievable rate is:

[0037]

[0038] Among them, the transmission from the access point AP to the base station is called the second hop. is the nth access point AP on sub-band j n The channel gain between the kth base station and α k，n Following the small-scale fading distribution, is large-scale fading, G t (n, k, j), G r (n, k, j) are the nth access point AP n To the kth base station BS k The transmission gain and receiving gain of the directional antenna, K = {1, 2, ..., K} is the set of base stations; the first term in the denominator of the signal-to-noise ratio is the interference in the same sub-band accumulated by all sensors related to other base stations in the same sub-band, σ 2 is the background noise variance, is the Doppler shift;

[0039] The available data rate depends on the smaller of the first and second hop data rates:

[0040]

[0041] Then, the transmission time when the sensor sends its task input in the uplink is:

[0042]

[0043] Among them, d m The amount of input data required to transfer program execution from the local sensor to the base station, Unload the variable for the task, indicating that the task Tm is offloaded from sensor m to the kth base station on subband j.

[0044] Preferably, the established base station computing offloading task model is specifically:

[0045] The execution time of the task on the base station is:

[0046]

[0047] Among them, {f mk , k∈K} is the computing resource allocation strategy, f mk [cycles / s]>0, for the kth base station BS k Assigned to the slave sensor m∈U k Unloaded tasks T m The amount of computing resources is the element in the matrix of computing resource allocation strategy;

[0048]

[0049] Among them, f k The computation rate is quantified as the computational resources provided by each base station and shared among the associated sensors, U k is the set of all sensors offloaded to the kth base station.

[0050] Preferably, in step S3, the specific contents of the time and energy consumption of the task calculations for obtaining the base station calculations and local calculations by the sensor through the mathematical model of data task offloading of each node are:

[0051] According to the local computing model of the sensor, the completion time and energy consumption of the local computing performed by the sensor are calculated;

[0052] According to the data task offloading model in the vehicle, the partial power and achievable rate of the sensor sending data to the base station in the first hop are calculated;

[0053] According to the offloading model of data tasks outside the vehicle, the feasible rate of the sensor sending data to the base station in the second hop is calculated. The transmission of the calculation task from the sensor to the AP is called the first hop, and the transmission from the AP to the base station is called the second hop.

[0054] obtaining an available data rate according to the first and second hop data rates, and calculating a transmission time when the sensor sends task input data in an uplink;

[0055] According to the base station computing offloading task model and the computing resource allocation strategy of the base station, the execution time of the task on the base station is calculated.

[0056] Preferably, in step S4, the overall optimization goal constructed is:

[0057]

[0058] in, is the weighted sum of all sensor offloading utilities, χ is the set of task offloading strategies, P is the power allocation matrix of each subnet access point AP, F is the base station computing resource allocation matrix, λ m ∈(,0], is the preference of the specified resource provider for sensor m;

[0059] The offloading objective function of sensor m is:

[0060]

[0061] Where 0≤k1≤1 indicates the preference of the specified sensor for processing delay when executing locally, and 1-k I is the preference of the energy consumption of the specified sensor when executing locally, k2 is the preference of the total delay experienced by sensor m when offloading the task, and 1-k2 is the preference of the energy consumption of sensor m due to the upload transmission;

[0062] The total delay experienced by sensor m when offloading the task is:

[0063]

[0064] The energy consumption of sensor m due to uploading transmission is:

[0065]

[0066] Preferably, the power allocation optimization target of each subnet access point AP is:

[0067]

[0068] Among them, θ mk To introduce new variables, an approximation of interference within the same sub-band accumulated for all sensors associated with other base stations on the same sub-band;

[0069] Based on solving the quasi-convex optimization problem, the optimized power allocation matrix P of each subnet AP is obtained. * ;

[0070] The optimization goal of computing resource allocation is:

[0071]

[0072] Based on solving the convex optimization problem, the optimized task offloading strategy set F is obtained. * ;

[0073] The optimization goal of task offloading allocation is:

[0074]

[0075] Among them, P * For a given access point power allocation matrix, F * Assign a matrix to a given computing resource;

[0076] Based on the heuristic task scheduling algorithm, the optimized base station computing resource allocation matrix χ is obtained.

[0077] Preferably, the specific content of step S6 is:

[0078] S61. Set the maximum number of optimizations and the convergence threshold, given the channel model, the amount of sensor offload data, the sensor local calculation frequency, the vehicle speed and other information;

[0079] S62 sets the power allocation matrix P of each subnet access point AP, the base station computing resource allocation matrix F, and the initial value of the task offloading strategy set χ;

[0080] S63. Use the binary search algorithm to search for the optimal value of the power allocation matrix P in the feasible domain and obtain P * , and update the public parameters;

[0081] S64. Optimize the base station calculation resource allocation matrix F under the new common parameters to obtain F * , and update the public parameters;

[0082] S65. Optimize the task offloading strategy set χ under the new common parameters, obtain χ℃, and update the common parameters;

[0083] S66. Repeat steps S62-S65 until the difference between the two optimizations is less than the convergence threshold or the maximum number of optimizations is reached to obtain the final P * 、F * and X * , which is the optimal task offloading strategy.

[0084] A low-complexity computing unloading system based on a high-speed rail vehicle-ground communication scenario, based on the low-complexity computing unloading method based on a high-speed rail vehicle-ground communication scenario, including a plurality of base stations, a plurality of carriage sub-networks, a wireless energy transmission acquisition module, a node data task unloading computing module, an optimization target decomposition computing module and a joint optimization module;

[0085] Each car sub-network includes an access node AP and multiple passive sensors. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station. Each sensor chooses to offload the computing task to a base station for computing or perform sensor local computing.

[0086] The wireless energy transmission acquisition module is used to collect energy by transmitting radio frequency signals through the subnet access node AP based on the energy transmission protocol, and uses a piecewise nonlinear model to obtain the receiving power of each sensor in each compartment;

[0087] The node data task offloading calculation module is used to calculate and obtain the time and energy consumption of the task calculation of the sensor for base station calculation and local calculation through the sensor local calculation model, the in-car data task offloading model, the out-car data task offloading model and the base station calculation offloading task model;

[0088] The optimization target decomposition calculation module is used to decompose the overall optimization target into the power allocation optimization target of each subnet access point AP, the base station computing resource allocation optimization target and the sensor task offloading decision optimization target, and solve the optimization variables of each optimization target;

[0089] The joint optimization module is used to alternately optimize and update each optimization variable through a joint optimization algorithm, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

[0090] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a low-complexity calculation offloading method and system based on the high-speed rail vehicle-ground communication scenario, which has the following advantages:

[0091] To solve the problem of limited sensor energy, the AP on the high-speed rail is used to supply energy to the sensor.

[0092] The interference of other sub-network sensors in the same sub-band in the multi-network environment and the Doppler effect of the actual high-speed rail environment are fully considered, and the constructed system model is more accurate;

[0093] By jointly designing the power allocation of each subnet access node, the base station computing resource allocation and the sensor task offloading decision, the performance of the wireless sensor task offloading computing system is optimized;

[0094] At the same time, it provides efficient and reliable communication means for multiple sensors, can be widely used in various fields such as the Internet of Vehicles, and has practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0096] Figure 1A schematic diagram of a low-complexity computation offloading method based on a high-speed rail vehicle-ground communication scenario provided by the present invention;

[0097] Figure 2 An overall schematic diagram of the low-complexity computing offloading system based on the high-speed rail train-to-ground communication scenario provided by the present invention. DETAILED DESCRIPTION

[0098] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0099] The embodiment of the present invention discloses a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario, comprising the following steps:

[0100] S1. Build an integrated system architecture based on high-speed rail data task offloading communication, including multiple base stations and carriage sub-networks. Each sub-network includes an access node AP and multiple passive sensors. In each sub-network, networking is achieved based on orthogonal frequency division multiplexing random access. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station.

[0101] S2. Determine the energy transmission protocol between the AP and the sensor within each subnet, use a piecewise nonlinear model to describe the wireless energy transmission process, obtain the receiving power of each sensor in each compartment, the subnet access node AP transmits the radio frequency signal, the sensor collects energy, and each sensor selects the calculation task to be offloaded to a base station for calculation or performs sensor local calculation;

[0102] S3. Establish a mathematical model for data task offloading at each node, including a sensor local calculation model, a car data task offloading model, a car data task offloading model, and a base station calculation offloading task model, and calculate the time and energy consumption of the sensor for base station calculation and local calculation through the mathematical model of data task offloading at each node;

[0103] S4. Construct an overall optimization objective function based on the weighted sum minimization of the time and energy consumed by the sensor task calculation in the system;

[0104] S5. Decompose the overall optimization goal into the power allocation optimization goal of each subnet access point AP, the base station computing resource allocation optimization goal and the sensor task offloading decision optimization goal, and solve the optimization variables of each optimization goal;

[0105] S6. Design a joint optimization algorithm to alternately optimize and update each optimization variable, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

[0106] In this embodiment, there are K base stations for remote communication with the train, and there are N high-speed rail carriages. The set of base stations is expressed as K = {1, 2, ..., K}, and the set of carriages is expressed as N = {1, 2, ..., N}. Each high-speed rail carriage includes an AP and m passive sensors without batteries. The set of all sensors is defined as U = {1, 2, ..., U}. All sensors can choose to offload computing tasks to a nearby base station or sensor for local computing. The AP periodically performs wireless energy transmission to activate the sensor. The sensor is awakened after receiving the energy signal and interacts with the AP after obtaining sufficient energy. When uploading data to the base station, the AP operates as a relay and works in full-duplex mode to shorten the time to upload to the base station.

[0107] In order to further implement the above technical solution, in step S2, the specific content of mathematically describing the wireless energy transmission process is:

[0108] Using the piecewise nonlinear energy harvesting model, the received power of the mth sensor in the nth compartment is calculated as:

[0109]

[0110] Among them, ε n，m ∈(0, 1] is the conversion factor, p n is the transmission power of the nth AP to the sensor in its compartment, h n，m is the channel gain between the nth AP and the mth sensor, P th is the peak power available to the mth sensor.

[0111] In this embodiment, the calculation of each sensor task is divided into two stages: a wireless energy collection stage and a data task processing stage. In the wireless energy collection stage, in each car, the AP first transmits a radio frequency signal and the sensor collects energy. In the multi-sensor vehicle-to-ground task offloading and communication system, the piecewise nonlinear energy harvesting model has high accuracy and efficient performance.

[0112] In order to further implement the above technical solution, the sensor local calculation model established in step S3 is specifically:

[0113] Each sensor has one computation task at a time, denoted as T m , that is, atomic tasks, cannot be divided into subtasks, and each computing task is characterized by two parameters <d m , c m >, where d m[bits] refers to the amount of input data required to transfer program execution from the local device to the base station, c m [cycles] refers to the workload, that is, the amount of computation required to complete a task. m and c m The value of can be obtained by carefully analyzing the task execution;

[0114] When sensor m performs a task locally, the time to complete the task is:

[0115]

[0116] In order to calculate the energy consumption of the sensor when performing tasks locally, the energy consumption model of each computing cycle ε = κf is adopted 2 , where κ is the energy coefficient that depends on the chip architecture, and f is the CPU frequency;

[0117] Sensor m performs task T locally m Energy consumption for:

[0118]

[0119] Among them, c m is the workload, i.e. the amount of computation required to complete the task, is the local computing power of the sensor, and κ is the energy coefficient that depends on the chip architecture.

[0120] In order to further implement the above technical solution, the in-car data task unloading model established in step S3 is specifically:

[0121] If sensor m sends its task T m If the data is offloaded to the base station, the delay incurred includes: (i) the time to send the input to the base station in the uplink; (ii) the time to perform the task at the base station; (iii) the time to send the output of the base station back to the user in the downlink. Since the size of the output is usually much smaller than the input, and the downlink data rate is much higher than the uplink data rate, the delay of the transmission downlink is ignored in the calculation.

[0122] The OFDM transmission protocol is used in the uplink data offloading process, and its operating band B is divided into M equal sub-bands with sizes of:

[0123] W=B / M

[0124] The size of M can ensure that each sensor associated with the same base station can be allocated to a non-overlapping sub-band, so each base station can serve at most M sensors at the same time, and the set of all served sensors of the same base station is defined as M = {1, 2, ..., M};

[0125] Task offloading variables Represents task T m The sensor m is offloaded to the kth base station BS on the subband k , the set G contains all the task offloading decision sets χ, U off =U k∈K U k is the set of all offload task sensors, where U k is the set of all sensors unloaded to the kth base station;

[0126] Since each task can be executed locally or offloaded to at most one base station, a feasible offloading strategy must satisfy the constraints:

[0127]

[0128] Since the offloaded data will still be forwarded by the AP, the transmission of the computing task from sensor m to AP is called the first hop, and the transmission from AP to the base station is called the second hop. It should be noted that since the AP operates in full-duplex mode, the task transmission on the first hop and the second hop is parallel. The AP receives data from sensor m and forwards the data to the base station on the same sub-band at the same time. The signal of the first hop will be affected by the background noise and self-interference SI cancellation. The residual SI value after cancellation can be expressed by the transmit power, which is convenient for calculation. Specifically:

[0129] The sensor is connected to the nth access point AP via sub-band j. n The first hop signal-to-noise ratio SINR is:

[0130]

[0131] in, is the nth access point AP on sub-band j n The channel gain between the mth sensor, G t (m, n, j), G r (m, n, j) are the distance from sensor m to the nth access point AP n The transmission gain and receiving gain of the directional antenna, βp n is the residual SI value, β is the SI cancellation level of the access point, which is a non-negative parameter, and p n The nth access point AP n The transmission power, that is, the transmission power of the wireless radio frequency radiated by the sensor in the car, σ 2 is the background noise variance;

[0132] According to the power splitting scheme, the sensor's received power is divided into two streams, specifically:

[0133] The power of the sensor to transmit uploaded data is:

[0134]

[0135] Among them, the remaining energy (1-α n，m ) n，m is offset by background noise and self-interference, α n，m ∈(0, 1] is a PS constant, P n，m is the received power of the mth sensor in the nth compartment, U is the set of sensors, and N is the set of compartments;

[0136] The sensor sends data to the kth base station BS at the first hop k The achievable rate is:

[0137]

[0138] Where W is the size of dividing the operating frequency band B into M equal sub-bands, It means that sensor m is connected to the nth access point AP via sub-band j. n The first hop signal-to-noise ratio is the transmission of the computing task from the sensor m to the access point AP;

[0139] The off-carriage data task offloading model established in step S3 is specifically:

[0140] Considering the same-channel interference, sensor m sends data to the kth base station BS in the second hop k The achievable rate is:

[0141]

[0142] The transmission from the access point AP to the base station is called the second hop. is the nth access point AP on sub-band j n The channel gain between the kth base station and α k，n Following the small-scale fading distribution, is large-scale fading, G t (n, k, j), G r (n, k, j) are the nth access point AP n To the kth base station BS k The transmission gain and receiving gain of the directional antenna are K = {1, 2, ..., K}, which is the set of base stations; the first term in the denominator of the signal-to-noise ratio is the interference in the same sub-band accumulated by all sensors related to other base stations in the same sub-band, that is, is the interference in the same sub-band accumulated by all sensors associated with other base stations on the same sub-band, where Unload variables for tasks, indicates that the task is offloaded from sensor a to the rth base station on sub-band j, pb is the transmission power of the bth AP uploading the task to the base station, is the channel gain between sensor a and the kth base station on sub-band j, σ 2 is the background noise variance, is the Doppler shift;

[0143]

[0144] where J0 is the zero-order Bessel function of the first kind, is the maximum Doppler shift, τ LOS =D(t) / c is the arrival time of the line-of-sight link, D(t) is the real-time distance, and c is the speed of light;

[0145] The available data rate depends on the smaller of the first and second hop data rates:

[0146]

[0147] Then, the transmission time when the sensor sends its task input in the uplink is:

[0148]

[0149] Among them, d m The amount of input data required to transfer program execution from the local sensor to the base station, Unload the variable for the task, indicating that the task T m is offloaded from sensor m to the kth base station on subband j.

[0150] In order to further implement the above technical solution, the established base station computing offloading task model is specifically as follows:

[0151] Each base station can provide computing offloading services to multiple sensors at the same time. After receiving the offloading task from the sensor, the base station will perform the task on behalf of the sensor and return the output result to the sensor after completion.

[0152] The execution time of the task on the base station is:

[0153]

[0154] Among them, {f mk , k∈K} is the computing resource allocation strategy, f mk [cycles / s]>0, for the kth base station BS k Assigned to the slave sensor m∈U k Unloaded tasks Tm The amount of computing resources is the element in the matrix of computing resource allocation strategy;

[0155] A feasible computing resource allocation strategy must satisfy the computing resource constraints:

[0156]

[0157] Among them, f k The computation rate is quantified as the computational resources provided by each base station and shared among the associated sensors, U k is the set of all sensors offloaded to the kth base station.

[0158] In order to further implement the above technical solution, in step S3, the specific contents of the time and energy consumption of the task calculation of the sensor for base station calculation and local calculation are calculated by the mathematical model of each node data task offloading:

[0159] According to the local computing model of the sensor, the completion time and energy consumption of the local computing performed by the sensor are calculated;

[0160] According to the data task offloading model in the vehicle, the partial power and achievable rate of the sensor sending data to the base station in the first hop are calculated;

[0161] According to the offloading model of data tasks outside the vehicle, the feasible rate of the sensor sending data to the base station in the second hop is calculated. The transmission of the calculation task from the sensor to the AP is called the first hop, and the transmission from the AP to the base station is called the second hop.

[0162] obtaining an available data rate according to the first and second hop data rates, and calculating a transmission time when the sensor sends task input data in an uplink;

[0163] According to the base station computing offloading task model and the computing resource allocation strategy of the base station, the execution time of the task on the base station is calculated.

[0164] In order to further implement the above technical solution, in step S4, the overall optimization goal constructed is:

[0165]

[0166] Among them, C1 and C2 mean that the data offloading task of the sensor can only be calculated locally or offloaded to the base station for calculation. C3 means that the data offloading task of the sensor can serve at most one sensor in each sub-band of each base station. C4 means that each base station must allocate a positive computing resource to each sensor. C5 means that the total computing resource allocated by each base station to all related sensors cannot exceed the computing capacity of the base station. C6 means that the power allocated by each access point to each sensor when it is working cannot be greater than the maximum threshold value. is the weighted sum of all sensor offloading utilities, χ is the set of task offloading strategies, P is the power allocation matrix of each subnet access point AP, F is the base station computing resource allocation matrix, λ m ∈(0, 1], is the preference of the specified resource provider for sensor m;

[0167] In this embodiment, according to the importance of the sensor, the resource provider can prioritize sensors with higher importance by improving their system performance. For example, the computation tasks of sensors involved in HSR safety operations should be prioritized to sensors with high values ​​of λ m ;

[0168] The offloading objective function of sensor m is:

[0169]

[0170] Where 0≤k1≤1 indicates the preference of the specified sensor for processing delay when executing locally, and 1-k I is the preference of the energy consumption of the specified sensor when executing locally, k2 is the preference of the total delay experienced by sensor m when offloading the task, and 1-k2 is the preference of the energy consumption of sensor m due to the upload transmission;

[0171] The total delay experienced by sensor m when offloading the task is:

[0172]

[0173] The energy consumption of sensor m due to uploading transmission is:

[0174]

[0175] Furthermore, the original problem with high complexity is transformed into an equivalent main problem and a set of sub-problems with lower complexity;

[0176] First, rewrite the optimization problem of the constructed overall optimization objective as follows:

[0177]

[0178] stc1-c6

[0179] By decoupling the offloading task constraints in c1-c3 and the resource allocation constraints in c4-c6, it is equivalent to solving the following offloading task OT problem:

[0180]

[0181] stc1-c3

[0182] Among them, J *(χ) is the optimal value function corresponding to the resource allocation RA problem, written as:

[0183]

[0184] stc4-c6

[0185] definition and and η m =k2λ m x mk c m , then the RA problem can be rewritten as:

[0186]

[0187] stc4-c6

[0188] Since the RA problem in the above formula has a separable structure, that is, the objectives and constraints corresponding to power allocation and computing resource allocation can be decoupled from each other, this feature is used to decouple the RA problem into two independent problems, namely power allocation PA and computing resource allocation CRA, and they are processed separately.

[0189] In order to further implement the above technical solution, the power allocation optimization target of each subnet access point AP is:

[0190]

[0191] Among them, θ mk To introduce new variables, an approximation of interference within the same sub-band accumulated for all sensors associated with other base stations on the same sub-band;

[0192] Based on solving the quasi-convex optimization problem, a low-complexity dichotomy method is used, which is greater than the set threshold, and the objective value of the optimization function is set to ξ(χ, P*), and the optimized power allocation matrix P* of each subnet AP is obtained;

[0193] The optimization goal of computing resource allocation is:

[0194]

[0195] Among them, the two constraints are convex and the second-order derivative of the objective function is always greater than 0, so the CRA problem is a convex optimization problem. Through the KKT condition, the objective value of the optimization function of the problem is set to ρ(χ, F*), and the optimized task offloading strategy set F* is obtained;

[0196] The optimization goal of task offloading allocation is:

[0197]

[0198] Wherein, P* is the given access point power allocation matrix, and F* is the given computing resource allocation matrix;

[0199] Based on the heuristic task scheduling algorithm, the optimized base station computing resource allocation matrix χ is obtained, specifically:

[0200] First, the algorithm starts with an empty set χ;

[0201] Then, for each sensor, select its objective function J * (χ) The minimum offloading decision, which puts all the sensor decisions into the set χ;

[0202] Finally, if there is a sensor decision Remove this decision from the set χ After that, make J * (χ)<(1-ζ)J * (χ) (where 1-ζ is the threshold set for each decrease), the relevant set is output; similarly, if there is a decision of a sensor In making decisions Exchange to set χ, which is specifically represented by removing the current decision of sensor m from set χ and the current decision of sensor v∈U is removed from the set χ Then, in order to ensure the constraints in the task offloading allocation optimization objective function, the decision Exchange to the set so that J * (χ)<(1-ζ)J * (χ), then output the related set χ.

[0203] In this embodiment, the main idea of ​​performing alternating optimization and updating on each optimization variable is: first, the remaining optimization variables are regarded as constants, the optimal value of a certain variable is solved and the common parameters are updated, and then the remaining variables are optimized in turn according to this method, and an alternating optimization is completed at this time; the above process is repeated until the preset number of optimizations is reached or the algorithm converges, and the final parameters of the desired joint optimization can be obtained.

[0204] In order to further implement the above technical solution, the specific content of step S6 is:

[0205] S61. Set the maximum number of optimizations and the convergence threshold, given the channel model, the amount of sensor offload data, the sensor local calculation frequency, the vehicle speed and other information;

[0206] S62 sets the power allocation matrix P of each subnet access point AP, the base station computing resource allocation matrix F, and the initial value of the task offloading strategy set χ;

[0207] S63. Use the binary search algorithm to search for the optimal value of the power allocation matrix P in the feasible domain and obtain P * , and update the public parameters;

[0208] S64. Optimize the base station calculation resource allocation matrix F under the new common parameters to obtain F * , and update the public parameters;

[0209] S65. Optimize the task offloading strategy set χ under the new common parameters, obtain χ*, and update the common parameters;

[0210] S66. Repeat steps S62-S65 until the difference between the two optimizations is less than the convergence threshold or the maximum number of optimizations is reached, and obtain the final P*, F* and χ*, which is the optimal task offloading strategy.

[0211] A low-complexity computing unloading system based on a high-speed rail vehicle-ground communication scenario, based on a low-complexity computing unloading method based on a high-speed rail vehicle-ground communication scenario, including several base stations, several carriage sub-networks, a wireless energy transmission acquisition module, a node data task unloading computing module, an optimization target decomposition computing module and a joint optimization module;

[0212] Each car sub-network includes an access node AP and multiple passive sensors. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station. Each sensor chooses to offload the computing task to a base station for computing or perform sensor local computing.

[0213] The wireless energy transmission acquisition module is used to collect energy by transmitting radio frequency signals through the subnet access node AP based on the energy transmission protocol, and uses a piecewise nonlinear model to obtain the receiving power of each sensor in each compartment;

[0214] The node data task offloading calculation module is used to calculate and obtain the time and energy consumption of the task calculation of the sensor for base station calculation and local calculation through the sensor local calculation model, the in-car data task offloading model, the out-car data task offloading model and the base station calculation offloading task model;

[0215] The optimization target decomposition calculation module is used to decompose the overall optimization target into the power allocation optimization target of each subnet access point AP, the base station computing resource allocation optimization target and the sensor task offloading decision optimization target, and solve the optimization variables of each optimization target;

[0216] The joint optimization module is used to alternately optimize and update each optimization variable through a joint optimization algorithm, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

[0217] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a low-complexity computation offloading method based on a high-speed rail vehicle-ground communication scenario.

[0218] A processing terminal includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is implemented.

[0219] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0220] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A low-complexity computation offloading method based on high-speed rail vehicle-ground communication scenario, characterized in that: The following steps are involved: S1. Build an integrated system architecture based on high-speed rail data task offloading communication, including multiple base stations and carriage sub-networks. Each sub-network includes an access node AP and multiple passive sensors. In each sub-network, networking is achieved based on orthogonal frequency division multiplexing random access. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station. S2. Determine the energy transmission protocol between the AP and the sensor within each subnet, use a piecewise nonlinear model to describe the wireless energy transmission process, obtain the receiving power of each sensor in each compartment, the subnet access node AP transmits the radio frequency signal, the sensor collects energy, and each sensor selects the calculation task to be offloaded to a base station for calculation or performs sensor local calculation; S3. Establish a mathematical model for data task offloading at each node, including a sensor local calculation model, a car data task offloading model, a car data task offloading model, and a base station calculation offloading task model, and calculate the time and energy consumption of the sensor for base station calculation and local calculation through the mathematical model of data task offloading at each node; S4. Construct an overall optimization objective function based on the weighted sum minimization of the time and energy consumed by the sensor task calculations in the system; S5. Decompose the overall optimization goal into the power allocation optimization goal of each subnet access point AP, the base station computing resource allocation optimization goal and the sensor task offloading decision optimization goal, and solve the optimization variables of each optimization goal; S6. Design a joint optimization algorithm to alternately optimize and update each optimization variable, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

2. According to claim 1, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is characterized in that: In step S2, the specific content of mathematically describing the wireless energy transmission process is: Using the piecewise nonlinear energy harvesting model, the received power of the mth sensor in the nth compartment is calculated as: Among them, ε n，m ∈(0, 1 ] is the conversion factor, p n is the transmission power of the nth AP to the sensor in its compartment, h n，m is the channel gain between the nth AP and the mth sensor, P th is the peak power available to the mth sensor.

3. According to claim 1, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is characterized in that: The sensor local calculation model established in step S3 is specifically: When sensor m performs a task locally, the time to complete the task is: Sensor m performs task T locally m The energy consumption is: Among them, c m is the workload, i.e. the amount of computation required to complete the task, is the local computing power of the sensor, and κ is the energy coefficient that depends on the chip architecture.

4. According to claim 1, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is characterized in that: The in-car data task offloading model established in step S3 is specifically: The power of the sensor to transmit uploaded data is: Among them, the remaining energy (1-α n,m ) n，m is offset by background noise and self-interference, α n，m ∈(0, 1] is a PS constant, P n，m is the received power of the mth sensor in the nth compartment, U is the set of sensors, and N is the set of compartments; The sensor sends data to the kth base station BS at the first hop k The achievable rate is: Where W is the size of dividing the operating frequency band B into M equal sub-bands, It means that sensor m is connected to the nth access point AP via sub-band j. n The first hop signal-to-noise ratio is the transmission of the computing task from the sensor m to the access point AP; in, is the nth access point AP on sub-band j n The channel gain between the mth sensor, G t (m, n, j), G r (m, n, j) are the distance from sensor m to the nth access point AP n The transmission gain and receiving gain of the directional antenna, βp n is the residual SI value, β is the SI cancellation level of the access point, which is a non-negative parameter, and p n The nth access point AP n The transmission power, that is, the transmission power of the wireless radio frequency radiated by the sensor in the car, σ 2 is the background noise variance; The off-carriage data task offloading model established in step S3 is specifically: Sensor m sends data to the kth base station BS at the second hop k The achievable rate is: The transmission from the access point AP to the base station is called the second hop. is the nth access point AP on sub-band j n The channel gain between the kth base station and α k，n Following the small-scale fading distribution, is large-scale fading, G t (n, k, j), G r (n, k, j) are the nth access point AP n To the kth base station BS k The transmission gain and receiving gain of the directional antenna, K = {1, 2, ..., K} is the set of base stations; the first term in the denominator of the signal-to-noise ratio is the interference in the same sub-band accumulated by all sensors related to other base stations in the same sub-band, σ 2 is the background noise variance, is the Doppler shift; The available data rate depends on the smaller of the first and second hop data rates: Then, the transmission time when the sensor sends its task input in the uplink is: Among them, d m The amount of input data required to transfer program execution from the local sensor to the base station, Unload the variable for the task, indicating that the task T m is offloaded from sensor m to the kth base station on subband j.

5. According to a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario according to claim 1, it is characterized in that: The established base station computing offloading task model is as follows: The execution time of the task on the base station is: Among them, {f mk , k∈K} is the computing resource allocation strategy, f mk [cycles / s]>0, for the kth base station BS k Assigned to the slave sensor m∈U k Unloaded tasks T m The amount of computing resources is the element in the matrix of computing resource allocation strategy; Among them, f k The computation rate is quantified as the computational resources provided by each base station and shared among the associated sensors, U k is the set of all sensors offloaded to the kth base station.

6. According to claim 1, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is characterized in that: In step S3, the specific contents of the task calculation time and energy consumption of the sensor for base station calculation and local calculation are obtained by calculating the mathematical model of data task offloading of each node: According to the local computing model of the sensor, the completion time and energy consumption of the local computing performed by the sensor are calculated; According to the data task offloading model in the vehicle, the partial power and achievable rate of the sensor sending data to the base station in the first hop are calculated; According to the offloading model of data tasks outside the vehicle, the feasible rate of the sensor sending data to the base station in the second hop is calculated. The transmission of the calculation task from the sensor to the AP is called the first hop, and the transmission from the AP to the base station is called the second hop. obtaining an available data rate according to the first and second hop data rates, and calculating a transmission time when the sensor sends task input data in an uplink; According to the base station computing offloading task model and the computing resource allocation strategy of the base station, the execution time of the task on the base station is calculated.

7. According to claim 1, a low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario is characterized in that: In step S4, the overall optimization goal constructed is: in, is the weighted sum of all sensor offloading utilities, χ is the set of task offloading strategies, P is the power allocation matrix of each subnet access point AP, F is the base station computing resource allocation matrix, λ m ∈(0, 1], is the preference of the specified resource provider for sensor m; The offloading objective function of sensor m is: Among them, 0≤k1≤1 represents the preference of the specified sensor for processing delay when executing locally, 1-k1 represents the preference of the specified sensor for energy consumption when executing locally, k2 represents the preference for the total delay experienced by sensor m when offloading the task, and 1-k2 represents the preference for the energy consumption caused by the upload transmission of sensor m; The total delay experienced by sensor m when offloading the task is: The energy consumption of sensor m due to uploading transmission is:

8. A low-complexity calculation offloading method based on high-speed rail vehicle-ground communication scenario according to claim 7, characterized in that: The power allocation optimization goal of each subnet access point AP is: Among them, θ mk To introduce new variables, an approximation of interference within the same sub-band accumulated for all sensors associated with other base stations on the same sub-band; Based on solving the quasi-convex optimization problem, the optimized power allocation matrix P of each subnet AP is obtained. * ; The optimization goal of computing resource allocation is: Based on solving the convex optimization problem, the optimized task offloading strategy set F is obtained. * ; The optimization goal of task offloading allocation is: Among them, P * For a given access point power allocation matrix, F * Assign a matrix to a given computing resource; Based on the heuristic task scheduling algorithm, the optimized base station computing resource allocation matrix χ is obtained.

9. A low-complexity calculation offloading method based on high-speed rail vehicle-ground communication scenario according to claim 8, characterized in that: The specific content of step S6 is: S61. Set the maximum number of optimizations and the convergence threshold, given the channel model, the amount of sensor offload data, the sensor local calculation frequency, the vehicle speed and other information; S62 sets the power allocation matrix P of each subnet access point AP, the base station computing resource allocation matrix F, and the initial value of the task offloading strategy set χ; S63. Use the binary search algorithm to search for the optimal value of the power allocation matrix P in the feasible domain and obtain P * , and update the public parameters; S64. Optimize the base station calculation resource allocation matrix F under the new common parameters to obtain F * , and update the public parameters; S65. Optimize the task offloading strategy set χ under the new common parameters to obtain χ ＊ , and update the public parameters; S66. Repeat steps S62-S65 until the difference between the two optimizations is less than the convergence threshold or the maximum number of optimizations is reached to obtain the final P ＊ 、F ＊ and χ ＊ , which is the optimal task offloading strategy.

10. A low-complexity computing offloading system based on high-speed rail vehicle-ground communication scenario, characterized in that: A low-complexity calculation offloading method based on a high-speed rail vehicle-ground communication scenario according to any one of claims 1 to 9, comprising a plurality of base stations, a plurality of carriage sub-networks, a wireless energy transmission acquisition module, a node data task offloading calculation module, an optimization target decomposition calculation module and a joint optimization module; Each car sub-network includes an access node AP and multiple passive sensors. Each sensor in each sub-network interacts with the sub-network access point AP and then establishes a connection with the base station. Each sensor chooses to offload the computing task to a base station for computing or perform sensor local computing. The wireless energy transmission acquisition module is used to collect energy by transmitting radio frequency signals through the subnet access node AP based on the energy transmission protocol, and uses a piecewise nonlinear model to obtain the receiving power of each sensor in each compartment; The node data task offloading calculation module is used to calculate and obtain the time and energy consumption of the task calculation of the sensor for base station calculation and local calculation through the sensor local calculation model, the in-car data task offloading model, the out-car data task offloading model and the base station calculation offloading task model; The optimization target decomposition calculation module is used to decompose the overall optimization target into the power allocation optimization target of each subnet access point AP, the base station computing resource allocation optimization target and the sensor task offloading decision optimization target, and solve the optimization variables of each optimization target; The joint optimization module is used to alternately optimize and update each optimization variable through a joint optimization algorithm, solve the optimal configuration of the system, and obtain the optimal task offloading strategy.

Images