Collaborative energy-saving computing migration method integrating RF energy harvesting

Abstract

The present invention belongs to the technical field of wireless communication networks, and discloses a collaborative energy-saving computing migration method that integrates radio frequency energy harvesting. By designing a layered fog computing-assisted data collection architecture, an optimization problem that minimizes the total energy consumption of the system is constructed based on the joint optimization considerations of computing migration decisions, uplink bandwidth resource allocation, downlink bandwidth resource allocation, and base station power partitioning. In order to effectively solve this optimization problem, a new evaluation index is designed by integrating the concept of penalty function, and a collaborative energy-saving computing migration algorithm based on adaptive particle swarm is proposed. The algorithm constructs a dynamically changing inertia weight and a linearly adjusted penalty factor, which can change the spatial distribution density of the particle community in real time during the iterative search process to generate the optimal computing migration strategy under tolerable penalties; further, in order to prevent particles from crossing the exploration range, a speed boundary limit is introduced, which can reduce the probability of generating invalid solutions and improve the search effectiveness.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication networks, and in particular to a collaborative energy-saving computing migration method integrating radio frequency energy harvesting. Background Art

[0002] Thanks to the development of a new era enabled by 5G, the intelligent Internet of Things (IoT), integrated with artificial intelligence (AI), has brought vast development opportunities to various industries. It is driving product upgrades in areas such as smart security, smart cities, and smart homes, and establishing a multi-tiered business ecosystem. However, this new development model often leads to increasing network usage and explosive growth in user data, making it difficult to balance the energy consumption costs of IoT devices with the diverse service demands of end users.

[0003] To improve system energy efficiency and build a green and energy-saving architecture, fog computing, a flexible and convenient computing model, is widely used. This model can efficiently couple different device interfaces to quickly analyze the interaction data of IoT applications. In particular, based on wireless sensing technology and computing migration technology, physical servers deployed at the edge can effectively integrate available environmental resources and split and schedule wireless transmission link resources to handle some energy-intensive user computing tasks, achieving good edge-end collaboration. Currently, there are two representative solutions:

[0004] The first type of solution uses a more optimal strategy to reduce terminal data processing energy consumption and improve system energy efficiency to a certain extent. For example, the paper [Resource allocation and computation offloading for multi-access edge computing with fronthaul and backhaul constraints] constructs an energy consumption minimization problem in a vertically heterogeneous mobile edge computing scenario and divides this non-convex problem into multiple subproblems to effectively solve the optimal computation migration decision and uplink and downlink bandwidth resource allocation. However, with the commercialization of the construction of data network connectivity, the battery consumption cost of IoT devices is extremely high, and regular replacement of this component is difficult and relatively arduous.

[0005] The second type of solution, such as the paper [Computation efficiency maximization in wireless-powered mobile edge computing networks], studies a partial computation migration mechanism based on nonlinear energy harvesting. It maximizes processing efficiency by jointly optimizing migration time, harvesting time, and local computation frequency. It uses node harvesting capabilities within a fixed unit time, or uses linear or nonlinear energy harvesting modes to divide task migration time and energy harvesting time into time slots, without considering model differences between devices. Furthermore, these modes set the system cycle to an estimated value, which often causes waiting time errors for users to receive feedback results. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention proposes a collaborative energy-saving computing migration method that integrates RF energy harvesting based on a separate energy harvesting architecture. Based on the joint optimization considerations of computing migration decisions, uplink bandwidth resource ratio, downlink bandwidth resource ratio and macro base station transmission power split ratio, an optimization problem is constructed to minimize the total system energy consumption required to process all tasks; for this optimization problem, an optimal solution method is proposed, which can converge to a fixed value at a relatively fast speed. The migration strategy has high solution accuracy, can obtain the lowest energy consumption value, and has good performance advantages.

[0007] The collaborative energy-saving computing migration method integrating radio frequency energy harvesting described in the present invention comprises the following steps:

[0008] Step 1: Design a layered network model consisting of a user layer and a fog node layer; the user layer includes several IoT devices, and the fog node layer includes several macro base stations and fog nodes;

[0009] Step 2: The IoT device randomly sends a computing task request to the fog node through the macro base station at a fixed time interval T;

[0010] Step 3: The fog node receives the computing task request from the IoT device and determines whether the computing task generated by the device needs to be migrated. If not, the IoT device performs local computing. If so, the fog node allocates certain bandwidth resources to the IoT device based on the energy-minimizing migration strategy. The IoT device divides the transmit power of the macro base station and collects energy through the macro base station. When the computing task is completed, the macro base station sends the processing result to the IoT device.

[0011] Step 4: Optimize the energy-minimizing migration strategy, construct an optimization problem to minimize the total system energy consumption required to process all tasks, and solve it optimally to obtain the migration decision.

[0012] Furthermore, the step 2 is specifically as follows:

[0013] Assume that the task request information sent by IoT device i∈{1, 2, ..., M} to the fog node layer is defined as (D i ,T i max ,P i max ), where D i represents the size of task data to be processed by IoT device i, T i max is the maximum tolerable delay of IoT device i, P i max Indicates the power threshold of device i in saturation state.

[0014] Furthermore, in step 3, when the fog node layer receives the request information, it will generate a corresponding migration strategy (α i ,β i ,γ i ,μ i ), where α i represents the task computing migration decision of IoT device i, when α i =1, it means that the task is processed on the local device; when α i =0, it means that the task is migrated to the fog node for processing; β i represents the proportion of uplink bandwidth resources allocated by fog nodes to IoT device i; γ i Indicates the downlink channel bandwidth resource ratio of device i; μ i is the ratio of device i to the base station's transmit power.

[0015] Furthermore, when the fog node determines that the computing task request is for local computing, the computing task requested by the user layer will be processed by the IoT device;

[0016] Define the local computing power of IoT device i as f i l , the resulting local computation delay is expressed as follows:

[0017]

[0018] Correspondingly, the local computing energy consumption after processing the request task of IoT device i is expressed as:

[0019]

[0020] Among them, P i l is the local computation power of device i.

[0021] Furthermore, when IoT device i chooses to migrate tasks to fog nodes for processing, it usually selects macro base stations as relay forwarding nodes. In this case, the total energy consumption is composed of uplink transmission energy consumption, fog node computing energy consumption, and downlink transmission energy consumption.

[0022] First, the uplink transmission rate of device i is expressed as follows:

[0023]

[0024] Among them, B u Indicates the size of uplink channel bandwidth resources, P i u represents the transmission power of IoT device i, Indicates the uplink channel gain when device i sends a computing task, represents the power spectral density of the background Gaussian white noise of the uplink channel, λ represents the channel capacity gap, and λ≥1;

[0025] According to the description of the uplink transmission rate, the uplink transmission delay and transmission energy consumption of IoT device i are expressed as:

[0026]

[0027]

[0028] Next, the fog node processes the computing task received from device i, and the fog node computing delay is expressed as:

[0029]

[0030] where f f is the computing power of the fog node, and f f >f i l ;

[0031] The computing energy consumption of fog nodes is expressed as follows:

[0032]

[0033] Among them, P f is the computing power of the fog node;

[0034] Finally, based on the RF signal emitted by the macro base station, the IoT device uses a built-in power divider to decompose the macro base station's transmit power, using it for information decoding and forwarding and RF energy collection. Here, the transmission signal on the channel where device i is located is defined as follows:

[0035]

[0036] in, For downlink transmission signals, the corresponding mathematical expectation satisfies P s Indicates the transmit power of the macro base station, is Gaussian white noise, represents the downlink channel gain when IoT device i receives feedback data;

[0037] Accordingly, according to Shannon's theorem, the downlink transmission rate corresponding to IoT device i during the information decoding and forwarding process is expressed as follows:

[0038]

[0039] Among them, B d Indicates the size of downlink channel bandwidth resources, σ i d Represents the power spectral density of environmental noise;

[0040] The downlink transmission delay of receiving decoded information is expressed as:

[0041]

[0042] Among them D i ′ is the data size of the computing task processing result fed back to device i by the fog node layer;

[0043] The corresponding downlink transmission energy consumption is expressed as follows:

[0044]

[0045] In particular, in the RF energy harvesting process, in order to better reflect the logical relationship between power and energy conversion, a nonlinear energy harvesting model is constructed based on the channel state of the device and the uncertain energy demand. The energy harvested by physical network device i is defined as:

[0046]

[0047] Among them, parameters a and b are related parameters for controlling the energy harvesting module. By changing the values of a and b, the inclination of the target energy consumption function can be effectively adjusted, and the upper limit of the harvested energy consumption can be controlled to a certain extent.

[0048] Based on the above description, the total energy consumption of the system to complete task processing of IoT device i can be expressed as follows:

[0049]

[0050] Furthermore, in step 4, based on the transmission characteristics of the wireless link, an optimization problem of minimizing the total energy consumption of the system is constructed, aiming to jointly optimize the calculation migration decision α i , uplink bandwidth resource ratio β i , downlink bandwidth resource ratio γ i and base station transmit power split ratio μ i , to improve the overall energy efficiency of the system, construct the following optimization problem P1:

[0051]

[0052] The constraints are as follows,

[0053] T i l +T i u +T i f +T i d ≤T i max ,(a)

[0054]

[0055]

[0056] 0≤μ i ≤1,(d)

[0057]

[0058] The optimization problem P1 is to minimize the total energy consumption of the system after processing all the computing tasks of IoT devices;

[0059] Constraint (a) indicates that the total time to complete the computing task of IoT device i cannot exceed its own maximum tolerable delay;

[0060] Constraints (b) and (c) indicate that the uplink bandwidth resources and downlink bandwidth resources allocated to IoT devices cannot exceed the total channel bandwidth resources;

[0061] Constraint (d) represents the split ratio of the base station transmit power, where μ i P s It is used for information decoding and forwarding, and the rest is used for RF energy collection;

[0062] Constraint (e) represents the computation migration decision of IoT device i’s task, and its value is 0 or 1, indicating migration processing and local processing, respectively.

[0063] In order to better solve the minimum energy consumption of the system, the idea of external penalty function is introduced and the following penalty term is constructed:

[0064]

[0065] Among them, δ is the penalty factor. Since the goal is to minimize the total energy consumption of the system, δ needs to be set to 0. η is the penalty term coefficient, which needs to satisfy η ≥ 1. Similarly, for other inequality constraints, the following penalty functions are constructed respectively:

[0066]

[0067]

[0068]

[0069]

[0070] in,

[0071] Finally, the original optimization problem is mapped to minimizing the sum of the total energy consumption of the system and the penalty term. The corresponding optimization problem P2 is expressed as follows:

[0072]

[0073] Furthermore, a collaborative energy-saving computing migration algorithm based on adaptive particle swarm is used to optimize problem P2, which specifically includes the following steps:

[0074] The target search space dimension of the particle community is defined as 4M. Assuming that the particle community consists of K particles, due to the information sharing behavior between particles, each particle will be influenced by its neighboring particles to perform self-learning and adjust its speed and position in a targeted manner to collaboratively improve the overall search efficiency and generate a better computational migration strategy. The following briefly describes the process using particle k as an example, where k∈{1,2,...,K};

[0075] The particle is first randomly encoded into a 4M-dimensional vector as follows:

[0076]

[0077] in, represents the initial position of particle k, represents the initial position of particle k in different dimensions, It means that Different positions obtained by splitting into four dimensions respectively;

[0078] Next, the initial flight velocity of particle k is defined as:

[0079] in The initial flying speed of particle k in different dimensions represented by ;

[0080] Considering that the above optimization variables are all in the range of 0 to 1, the initial flight speed is set to a random number between [0, 1] and the speed is updated according to the following formula:

[0081]

[0082] Where m represents the m-th dimension vector, and m∈{1,2,...,4M}; n represents the n-th iteration, and n≥0; ω n is the inertia weight of the nth iteration, which indicates the degree of trust of the individual particle in the previous search action; c1 and c2 are the individual learning factor and the community learning factor, respectively, indicating the weight ratio of the next search action to adopt the individual learning experience and the shared experience of neighbors; r1 and r2 are random probabilities in the range of 0 to 1 that conform to the uniform distribution; represents the local optimal position of the m-th dimension vector searched by particle k in the n-th iteration; It represents the global optimal position of the m-th dimension vector searched by the entire particle community; the updated particle position is expressed as:

[0083] in Indicates the particle position after the nth iteration;

[0084] Finally, in the migration scenario of integrated RF energy harvesting, energy consumption is the main evaluation indicator for measuring system effectiveness. Combined with the above definitions of particle position and flight speed, the fitness function, a key element in the particle swarm algorithm, is expressed as the sum of energy consumption and penalty function, as follows:

[0085]

[0086] Where X is a 4M-dimensional vector consisting of the target solution strategy;

[0087] To ensure the search accuracy of the migration strategy and improve the inspiration of the particle motion state, the idea of dynamic inertia weight is adopted, as follows:

[0088]

[0089] Among them, ω max Indicates the upper limit of the inertia weight, ω min Indicates the lower limit of the inertia weight, n indicates the current nth iteration, and N indicates the maximum number of iterations of the algorithm;

[0090] At the same time, to avoid the particle position To prevent the overflow phenomenon, a boundary limit is set for the particle flight speed, which is updated as follows:

[0091]

[0092] Among them, v max is the upper limit of the particle flight speed, is the particle velocity of particle k in the mth dimension after the n+1th round update;

[0093] According to the search state of the particle at the previous moment To minimize the fitness function f(X k ) as the goal, and find the local optimal solution at the next moment and the global optimal solution The specific expressions are as follows:

[0094]

[0095]

[0096] Next, the particle's flight speed and position are iteratively updated using the currently explored optimal solution space until the iteration condition is no longer met, that is, when the maximum number of iterations N is reached, the optimal calculation migration strategy is output. In particular, a penalty amplification coefficient τ is introduced to make the penalty factor δ show a linear growth trend, so as to achieve the randomness of the early migration decision exploration and ensure the accuracy and stability of the later resource allocation strategy output. It is specifically expressed as:

[0097] δ n+1 =τδ n , where δ n+1 Expressed as the penalty factor at the n+1th iteration, δ n It is expressed as the penalty factor at the nth iteration.

[0098] The beneficial effects of the present invention are as follows: (1) Based on the joint optimization considerations of computational migration decisions, uplink bandwidth resource ratio, downlink bandwidth resource ratio, and macro base station transmit power split ratio, the present invention constructs an optimization problem for minimizing the total system energy consumption required to process all tasks; at the same time, in order to characterize the upper limit of energy that can be collected by different devices, the nonlinear energy collection mode is integrated and considered, and the tilt degree of the objective function can be effectively adjusted according to the circuit adjustment parameters and the power threshold value in the saturation state;

[0099] (2) In order to effectively solve the optimization problem of minimizing the total energy consumption of the system required to process all tasks, the present invention proposes a collaborative energy-saving computing migration algorithm based on an adaptive particle swarm. The algorithm introduces the idea of dynamic inertia weights, which can change the spatial distribution density of the particle swarm in real time during the iterative search process to better guide the particles to explore the solution space and improve the accuracy of the strategy solution. At the same time, the concept of penalty amplification coefficient is integrated to linearly adjust the penalty factor one by one to generate the optimal computing migration strategy under tolerable penalties, further improving the prediction effect of the results of precision or generalization. In particular, a flight speed upper limit is designed during the exploration process, which can constrain the solution space that exceeds the range to the boundary value, avoid excessive offset, reduce the probability of generating invalid solutions, and improve the search effectiveness.

[0100] (3) Compared with other solutions, the proposed algorithm consistently converges to a fixed value at a faster rate, and compared with the traditional particle swarm optimization algorithm, the solution accuracy of the migration strategy is improved by 7.36%. Furthermore, as the pressure of task data requests increases, the algorithm consistently achieves the lowest energy consumption value, demonstrating excellent performance advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 This is a fog computing migration model designed by the present invention that integrates radio frequency energy harvesting;

[0102] Figure 2 is the convergence of the fitness function of the method of the present invention under different inertia weights;

[0103] Figure 3 This is a comparison chart of the total system energy consumption of the method in the present invention and other methods under different task data sizes;

[0104] Figure 4 This is a comparison chart of the total energy consumption of the system with different fog node computing capabilities using the method in the present invention and other methods. DETAILED DESCRIPTION

[0105] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0106] like Figure 1 As shown, a collaborative energy-saving computing migration method integrating radio frequency energy harvesting includes the following steps:

[0107] (1) Design a layered network model, which consists of two parts: the user layer and the fog node layer. The user layer is mainly composed of IoT devices, and the fog node layer is mainly composed of fog nodes and macro base stations.

[0108] (2) There are M IoT devices deployed in the user layer, and every fixed time interval T, they will randomly send computing task requests to the fog node layer.

[0109] (3) In the fog node layer, the macro base station is mainly responsible for ensuring uninterrupted communication between IoT devices and fog nodes, and providing continuous availability energy supply services, so that connected IoT devices can flexibly and conveniently store additional energy to improve their self-powered capacity, extend their service life and reduce energy consumption. The fog node senses the IoT devices within its communication coverage range and determines whether the computing tasks generated by them need to be migrated. Based on the energy-minimizing migration strategy, it allocates a certain amount of bandwidth resources to the relevant devices. When the computing tasks are completed, the processing results are sent to the IoT devices.

[0110] Furthermore, step (2) specifically includes the following contents:

[0111] Assume that the task request information sent by IoT device i∈{1, 2, ..., M} to the fog node layer is defined as (D i ,T i max ,P i max ), where D i represents the size of task data to be processed by IoT device i, T i max is the maximum tolerable delay of IoT device i, P i max Indicates the power threshold of device i in saturation state. This value is usually determined by the device manufacturer and device model. When the fog node layer receives the request information, it will generate the corresponding migration strategy (α) with the goal of minimizing total energy consumption. i ,β i ,γ i ,μ i ), where α i represents the task computing migration decision of IoT device i, when α i =1, it means that the task is processed on the local device; when α i =0, it means that the task is migrated to the fog node for processing; β i represents the proportion of uplink bandwidth resources allocated by fog nodes to IoT device i; γ i Indicates the downlink channel bandwidth resource ratio of device i; μ i is the ratio of device i to the base station's transmit power.

[0112] Furthermore, step (3) specifically includes the following contents:

[0113] For local computing, the computing tasks requested by the user layer will be processed by the IoT device. The local computing capacity of IoT device i is defined as f i l , the resulting local computation delay is expressed as follows:

[0114]

[0115] Correspondingly, the local computing energy consumption after processing the request task of IoT device i is expressed as:

[0116]

[0117] Among them, P i l is the local computation power of device i.

[0118] When IoT device i chooses to migrate tasks to fog nodes for processing, it usually selects macro base stations as relay forwarding nodes. At this time, the total energy consumption is mainly composed of uplink transmission energy consumption, fog node computing energy consumption and downlink transmission energy consumption.

[0119] First, the uplink transmission rate of device i is expressed as follows:

[0120]

[0121] Among them, B u Indicates the size of uplink channel bandwidth resources, P i u represents the transmission power of IoT device i, Indicates the uplink channel gain when device i sends a computing task, represents the power spectral density of the background Gaussian white noise of the uplink channel, λ represents the channel capacity gap, and λ≥1;

[0122] According to the description of the uplink transmission rate, the uplink transmission delay and transmission energy consumption of IoT device i can be expressed as:

[0123]

[0124]

[0125] Next, the fog node processes the computing task received from device i. Similar to the local computing model, the fog node computing latency can be expressed as:

[0126]

[0127] where f f is the computing power of the fog node, and f f >f il ;

[0128] The computing energy consumption of fog nodes can be expressed as follows:

[0129]

[0130] Among them, P f is the computing power of the fog node;

[0131] Finally, based on the RF signal emitted by the macro base station, the IoT device can use the built-in power divider to decompose the macro base station's transmit power, using it for information decoding and forwarding and RF energy collection. Here, the transmission signal on the channel where device i is located is defined as follows:

[0132]

[0133] in, For downlink transmission signals, the corresponding mathematical expectation satisfies is Gaussian white noise.

[0134] Accordingly, according to Shannon's theorem, the downlink transmission rate corresponding to IoT device i during the information decoding and forwarding process can be expressed as follows:

[0135]

[0136] Among them, B d Indicates the size of downlink channel bandwidth resources, P s Indicates the transmit power of the macro base station, represents the downlink channel gain when IoT device i receives feedback data, Represents the power spectral density of the ambient noise.

[0137] The downlink transmission delay of receiving decoded information can be expressed as:

[0138]

[0139] Among them D i ′ is the data size of the computing task processing result fed back to device i by the fog node layer.

[0140] The corresponding downlink transmission energy consumption can be expressed as follows:

[0141]

[0142] In particular, in the RF energy harvesting process, in order to better reflect the logical relationship between power and energy conversion, a nonlinear energy harvesting model is constructed based on the channel state of the device and the uncertain energy demand. The energy harvested by physical network device i is defined as:

[0143]

[0144] Parameters a and b are related to controlling the energy harvesting module. By changing the values of a and b, the tilt of the target energy consumption function can be effectively adjusted, and the upper limit of harvested energy consumption can be controlled to a certain extent. Based on the above description, the total system energy consumption for completing task processing of IoT device i can be expressed as follows:

[0145]

[0146] Based on the transmission characteristics of wireless links, an optimization problem to minimize the total energy consumption of the system is constructed, aiming to jointly optimize the calculation migration decision α i , uplink bandwidth resource ratio β i , downlink bandwidth resource ratio γ i and base station transmit power split ratio μ i , to improve the overall energy efficiency of the system, construct the following optimization problem P1:

[0147]

[0148] The constraints are as follows,

[0149] T i l +T i u +T i f +T i d ≤T i max ,(a)

[0150]

[0151]

[0152] 0≤μ i ≤1,(d)

[0153]

[0154] The optimization problem P1 is to minimize the total energy consumption of the system after processing all the computing tasks of IoT devices;

[0155] Constraint (a) indicates that the total time to complete the computing task of IoT device i cannot exceed its own maximum tolerable delay;

[0156] Constraints (b) and (c) indicate that the uplink bandwidth resources and downlink bandwidth resources allocated to IoT devices cannot exceed the total channel bandwidth resources;

[0157] Constraint (d) represents the split ratio of the base station transmit power, where μ i P s It is used for information decoding and forwarding, and the rest is used for RF energy collection;

[0158] Constraint (e) represents the computation migration decision of IoT device i’s task, and its value is 0 or 1, indicating migration processing and local processing, respectively.

[0159] In order to better solve the minimum energy consumption of the system, the idea of external penalty function is introduced and the following penalty term is constructed:

[0160]

[0161] Among them, δ is the penalty factor. Since the goal is to minimize the total energy consumption of the system, δ needs to be set to 0. η is the penalty term coefficient, which needs to satisfy η ≥ 1. Similarly, for other inequality constraints, the following penalty functions are constructed respectively:

[0162]

[0163]

[0164]

[0165]

[0166] in,

[0167] Finally, the original optimization problem can be mapped to minimizing the sum of the total energy consumption of the system and the penalty term. The corresponding optimization problem P2 can be expressed as follows:

[0168]

[0169] The collaborative energy-saving computing migration algorithm based on adaptive particle swarm is used to optimize problem P2, which includes the following steps:

[0170] The target search space dimension of the particle community is defined as 4M. Assuming that the particle community consists of K particles, due to the information sharing behavior between particles, each particle will be influenced by its neighboring particles to perform self-learning and adjust its speed and position in a targeted manner to collaboratively improve the overall search efficiency and generate a better computational migration strategy. The following briefly describes the process using particle k as an example, where k∈{1, 2, ..., K}.

[0171] The particle is first randomly encoded into a 4M-dimensional vector as follows:

[0172]

[0173] in, represents the initial position of particle k, represents the initial position of particle k in different dimensions, It means that Different positions obtained by splitting into four dimensions.

[0174] Then, the initial flight velocity of particle k can be defined as:

[0175]

[0176] Considering that the above optimization variables are all in the range of 0 to 1, the initial flight speed is set to a random number between [0, 1] and the speed is updated according to the following formula:

[0177]

[0178] Where m represents the m-th dimension vector, and m∈{1, 2, ..., M}; n represents the n-th iteration, and n≥0; ω n is the inertia weight of the nth iteration, which indicates the degree of trust of the individual particle in the previous search action; c1 and c2 are the individual learning factor and the community learning factor, respectively, indicating the weight ratio of the next search action to adopt the individual learning experience and the shared experience of neighbors; r1 and r2 are random probabilities in the range of 0 to 1 that conform to the uniform distribution; represents the local optimal position of the m-th dimension vector searched by particle k in the n-th iteration; This update method enables particle k to better search the solution space based on the collaborative sharing information of neighboring particles while maintaining its own velocity trend, and the updated particle position can be expressed as:

[0179] in Represents the particle position after the nth iteration.

[0180] In addition, since energy consumption is the main evaluation indicator for measuring system effectiveness in the migration scenario of integrated RF energy harvesting, combined with the above definitions of particle position and flight speed, the fitness function, a key element in the particle swarm algorithm, can be expressed as the sum of energy consumption and penalty function, as follows:

[0181]

[0182] Where X is a 4M-dimensional vector consisting of the target solution strategy.

[0183] To ensure the search accuracy of the migration strategy and improve the inspiration of the particle motion state, the idea of dynamic inertia weight is adopted, as follows:

[0184]

[0185] Among them, ω max Indicates the upper limit of the inertia weight, ω min represents the lower limit of the inertia weight, n represents the current nth iteration, and N represents the maximum number of iterations of the algorithm. Therefore, the value of the inertia weight decreases with the increase of the number of iterations, which can effectively control the actual convergence performance and ensure that the particle swarm is getting closer and closer to the ideal solution.

[0186] At the same time, to avoid the particle position To prevent the overflow phenomenon, a boundary limit is set for the particle flight speed, which is updated as follows:

[0187]

[0188] Among them, v max is the upper limit of the particle flight speed, It is the particle velocity of particle k in the mth dimension after the n+1th round of update.

[0189] According to the search state of the particle at the previous moment To minimize the fitness function f(X k ) as the goal, we can find the local optimal solution at the next moment and the global optimal solution The specific expressions are as follows:

[0190]

[0191]

[0192] Next, the particle's flight speed and position can be iteratively updated using the currently explored optimal solution space until the iteration condition is no longer met, that is, when the maximum number of iterations N is reached, the optimal computational migration strategy is output. In particular, a penalty amplification coefficient τ is introduced to make the penalty factor δ increase linearly, thereby achieving randomness in the early stage of migration decision exploration and ensuring the accuracy and stability of the later resource allocation strategy output. Specifically, it is expressed as:

[0193] δ n+1 =τδ n , where δ n+1 Expressed as the penalty factor at the n+1th iteration, δ n It is expressed as the penalty factor at the nth iteration.

[0194] The specific implementation of the method of the present invention verifies that the proposed algorithm can achieve higher convergence efficiency and solution accuracy.

[0195] In the simulation experiment, it is assumed that there are 5 IoT devices deployed in the user layer to interact with the fog node layer for task and energy collaborative perception. The relevant experimental parameters are set as follows: The task data size D requested by IoT device i i Randomly generate between 10Kb and 100Kb, local calculation power P i l and uplink transmission power P i u The values of are between 1W and 5W. The macro base station transmission power and fog node computing power at the fog node layer are 20W and 15W respectively. The power threshold value P in the saturation state of the system circuit is i max Randomly generate between 50,000 and 100,000, local computing power. Randomly generate between 20Kb / s and 40Kb / s. In particular, to facilitate simple calculation, define the data size D of the returned task result i ' is the requested task data size D i half of the maximum tolerable delay T for task processing i max is 0.5s, the computing power of the fog node is f f The value is 2Mb / s, and the uplink channel bandwidth resource size is B u and downlink channel bandwidth resource size B d 25Kb / s and 15Kb / s respectively, uplink channel gain and downlink channel gain 10 respectively -5 and 10 -6 , the actual channel difference parameter λ is set to 5, and the Gaussian white noise power spectral density of the uplink and downlink channels is and Set to 10 respectively -9 and 10 -10 , adjust the relevant parameters a and b of the nonlinear energy harvesting module to 10 and 2 respectively.

[0196] like Figure 2 As shown in the figure, when the inertia weight is set to 0, the fitness function remains unchanged and is a constant value, and this value is the minimum value corresponding to the initial flight speed. This is because the particle community has not produced any evolution process, that is, the particles have not diffused and cannot seek the optimal solution; when the inertia weight is 1, each particle can explore a wider spatial domain in the early stage, so that the fitness function can reach the optimal solution more quickly, but its search efficiency and convergence accuracy in the later iteration are relatively low.

[0197] In addition, if Figure 3 As shown in the figure, the energy consumption under all schemes increases with the increase of task data size, among which "Full local" means that all computing tasks of IoT devices are processed on the device itself; "Full offloading" means that all computing tasks are migrated to a unique fog node for processing; "PSO" represents a computing migration scheme based on the traditional particle swarm algorithm; "APSO-CESCO" represents the computing migration scheme proposed in the present invention. Since the computing power of IoT devices is far less than that of fog nodes, the total energy consumption under full local processing is the highest, followed by the full migration scheme; furthermore, since the traditional particle swarm scheme integrates RF energy harvesting technology and implements joint optimization of migration decisions, uplink and downlink bandwidth resources and macro base station transmission power split ratios, the total energy consumption of the system can be further reduced; however, the search accuracy under this scheme is low, and it is easy to miss a better resource allocation strategy. The inertia weight and penalty function ideas introduced in the invention can improve the solution accuracy and obtain the lowest system energy consumption. Not only that, as Figure 4 As shown in the figure, if the computing power of the fog node is small, the total energy consumption under the full migration processing scheme is the highest. This is because reducing the communication overhead at this time is difficult to make up for the disadvantage of computing overhead. However, as the computing power of the node increases, its energy consumption value can be gradually reduced, that is, this disadvantage can be effectively alleviated. At the same time, since the local computing energy consumption is only closely related to the size of the user request data, local computing power and local computing power, it will not be affected by the changing computing power of the fog node. Therefore, the total energy consumption of the system under the full local processing scheme is a fixed constant. In addition, the scheme proposed in the present invention has a better particle space exploration mode than the PSO scheme based on fixed inertia weight, which makes the average energy consumption value converged to lower.

[0198] The above description is only a preferred embodiment of the present invention and is not intended to further limit the present invention. All equivalent changes made using the contents of the present invention description and drawings are within the scope of protection of the present invention.

Claims

1. A collaborative energy-saving computing migration method integrating radio frequency energy harvesting, characterized in that: The method steps are: Step 1: Design a layered network model consisting of a user layer and a fog node layer; the user layer includes several IoT devices, and the fog node layer includes several macro base stations and fog nodes; Step 2: The IoT device randomly sends a computing task request to the fog node through the macro base station at a fixed time interval T; Step 3: The fog node receives the computing task request from the IoT device and determines whether the computing task generated by it needs to be migrated; If the judgment is no, the IoT device performs local calculations; If the answer is yes, the fog node allocates certain bandwidth resources to the IoT device according to the energy consumption minimization migration strategy; The IoT device divides the transmit power of the macro base station, collects energy through the macro base station, and when the calculation task is completed, the macro base station sends the processing result to the IoT device; Step 4: Optimize the energy-minimizing migration strategy. Construct an optimization problem to minimize the total system energy consumption required to process all tasks, and find the optimal solution to obtain the migration decision. Specifically, Based on the transmission characteristics of wireless links, an optimization problem to minimize the total energy consumption of the system is constructed, aiming to jointly optimize the calculation migration decision α i , uplink bandwidth resource ratio β i , downlink bandwidth resource ratio γ i and base station transmit power split ratio μ i , to improve the overall energy efficiency of the system, construct the following optimization problem P1: The constraints are as follows, T i l +T i u +T i f +T i d ≤T i max , (a) 0≤μ i ≤1, (d) Among them, IoT device i∈{1, 2, ..., M}, E i The total energy consumption of the system to complete the task processing of IoT device i is T i l is the local computing delay of IoT device i, T i u is the uplink transmission delay of IoT device i, T i f Calculate the delay for the fog node, T i d The downlink transmission delay for receiving decoded information; The optimization problem P1 is to minimize the total energy consumption of the system after processing all the computing tasks of IoT devices; Constraint (a) indicates that the total time to complete the computing task of IoT device i cannot exceed its own maximum tolerable delay; Constraints (b) and (c) indicate that the uplink bandwidth resources and downlink bandwidth resources allocated to IoT devices cannot exceed the total channel bandwidth resources; Constraint (d) represents the split ratio of the base station transmit power, where μ i P s It is used for information decoding and forwarding, and the rest is used for RF energy collection; Constraint (e) represents the computation migration decision of IoT device i’s task, and its value is 0 or 1, indicating migration processing and local processing, respectively. In order to better solve the minimum energy consumption of the system, the idea of external penalty function is introduced and the following penalty term is constructed: Among them, δ is the penalty factor. Since the goal is to minimize the total energy consumption of the system, δ needs to be set to 0. η is the penalty term coefficient, which needs to satisfy η ≥ 1. Similarly, for other inequality constraints, the following penalty functions are constructed respectively: in, Finally, the original optimization problem is mapped to minimizing the sum of the total energy consumption of the system and the penalty term. The corresponding optimization problem P2 is expressed as follows: A collaborative energy-saving computing migration algorithm based on adaptive particle swarm is used to optimize problem P2.

2. The collaborative energy-saving computing migration method integrating radio frequency energy harvesting according to claim 1 is characterized in that: The step 2 is specifically as follows: Assume that the task request information sent by IoT device i∈{1, 2, ..., M} to the fog node layer is defined as (D i ,T i max ,P i max ), where D i represents the size of task data to be processed by IoT device i, T i max is the maximum tolerable delay of IoT device i, P i max Indicates the power threshold of device i in saturation state.

3. The collaborative energy-saving computing migration method integrating radio frequency energy harvesting according to claim 2 is characterized in that: In step 3, when the fog node layer receives the request information, it will generate the corresponding migration strategy (α i ,β i ,γ i ,μ i ), where α i represents the task computing migration decision of IoT device i, when α i =1, it means that the task is processed on the local device; when α i =0, it means that the task is migrated to the fog node for processing; β i represents the proportion of uplink bandwidth resources allocated by fog nodes to IoT device i; γ i Indicates the downlink channel bandwidth resource ratio of device i; μ i is the ratio of device i to the base station's transmit power.

4. The collaborative energy-saving computing migration method integrating radio frequency energy harvesting according to claim 3 is characterized in that: When the fog node determines that the computing task request is for local computing, the computing task requested by the user layer will be processed by the IoT device; Define the local computing power of IoT device i as f i l , the resulting local computation delay is expressed as follows: Correspondingly, the local computing energy consumption after processing the request task of IoT device i is expressed as: Among them, P i l is the local computation power of device i.

5. The collaborative energy-saving computing migration method integrating radio frequency energy harvesting according to claim 4 is characterized in that: When IoT device i chooses to migrate tasks to fog nodes for processing, it usually selects macro base stations as relay forwarding nodes. In this case, the total energy consumption is composed of three parts: uplink transmission energy consumption, fog node computing energy consumption, and downlink transmission energy consumption. First, the uplink transmission rate of device i is expressed as follows: Among them, B u Indicates the size of uplink channel bandwidth resources, P i u represents the transmission power of IoT device i, Indicates the uplink channel gain when device i sends a computing task, represents the power spectral density of the background Gaussian white noise of the uplink channel, λ represents the channel capacity gap, and λ≥1; According to the description of the uplink transmission rate, the uplink transmission delay and transmission energy consumption of IoT device i are expressed as: Next, the fog node processes the computing task received from device i, and the fog node computing delay is expressed as: where f f is the computing power of the fog node, and f f >f i l; The computing energy consumption of fog nodes is expressed as follows: Among them, P f is the computing power of the fog node; Finally, based on the RF signal emitted by the macro base station, the IoT device uses a built-in power divider to decompose the macro base station's transmit power, using it for information decoding and forwarding and RF energy collection. Here, the transmission signal on the channel where device i is located is defined as follows: in, For downlink transmission signals, the corresponding mathematical expectation satisfies P s Indicates the transmit power of the macro base station, is Gaussian white noise, represents the downlink channel gain when IoT device i receives feedback data; Accordingly, according to Shannon's theorem, the downlink transmission rate corresponding to IoT device i during the information decoding and forwarding process is expressed as follows: Among them, B d Indicates the size of downlink channel bandwidth resources, Represents the power spectral density of environmental noise; The downlink transmission delay of receiving decoded information is expressed as: Among them D i ′ is the data size of the computing task processing result fed back to device i by the fog node layer; The corresponding downlink transmission energy consumption is expressed as follows: In particular, in the RF energy harvesting process, in order to better reflect the logical relationship between power and energy conversion, a nonlinear energy harvesting model is constructed based on the channel state of the device and the uncertain energy demand. The energy harvested by physical network device i is defined as: Among them, parameters a and b are related parameters for controlling the energy harvesting module; In summary, the total energy consumption of the system to complete task processing of IoT device i is expressed as follows:

6. The collaborative energy-saving computing migration method integrating radio frequency energy harvesting according to claim 5 is characterized in that: The collaborative energy-saving computing migration algorithm based on adaptive particle swarm is used to optimize problem P2, which includes the following steps: The target search space dimension of the particle community is defined as 4M. Assuming that the particle community consists of K particles, due to the information sharing behavior between particles, each particle will be influenced by its neighboring particles to perform self-learning and adjust its speed and position in a targeted manner to collaboratively improve the overall search efficiency and generate a better computational migration strategy. The following briefly describes the process using particle k as an example, where k∈{1,2,...,K}; The particle is first randomly encoded into a 4M-dimensional vector as follows: in, represents the initial position of particle k, represents the initial position of particle k in different dimensions, It means that Different positions obtained by splitting into four dimensions respectively; Next, the initial flight velocity of particle k is defined as: in The initial flight speed of particle k in different dimensions represented by ; Considering that the optimization variables, i.e., the initial position and initial flight speed of particle k, are both in the range of 0 to 1, the initial flight speed is set to a random number between [0, 1], and the speed is updated according to the following formula: Where m represents the m-th dimension vector, and m∈{1,2,...,4M}; n represents the n-th iteration, and n≥0; ω n is the inertia weight of the nth iteration, which indicates the degree of trust of the individual particle in the previous search action; c1 and c2 are the individual learning factor and the community learning factor, respectively, indicating the weight ratio of the next search action to adopt the individual learning experience and the shared experience of neighbors; r1 and r2 are random probabilities in the range of 0 to 1 that conform to the uniform distribution; represents the local optimal position of the m-th dimension vector searched by particle k in the n-th iteration; It represents the global optimal position of the m-th dimension vector searched by the entire particle community; the updated particle position is expressed as: in Indicates the particle position after the nth iteration; Finally, in the migration scenario of integrated RF energy harvesting, energy consumption is the main evaluation indicator for measuring system effectiveness. Combined with the above definitions of particle position and flight speed, the fitness function, a key element in the particle swarm algorithm, is expressed as the sum of energy consumption and penalty function, as follows: Where X is a 4M-dimensional vector consisting of the target solution strategy; To ensure the search accuracy of the migration strategy and improve the inspiration of the particle motion state, the idea of dynamic inertia weight is adopted, as follows: Among them, ω max Indicates the upper limit of the inertia weight, ω min Indicates the lower limit of the inertia weight, n indicates the current nth iteration, and N indicates the maximum number of iterations of the algorithm; At the same time, to avoid the particle position To prevent the overflow phenomenon, a boundary limit is set for the particle flight speed, which is updated as follows: Among them, v max is the upper limit of the particle flight speed, is the particle velocity of particle k in the mth dimension after the n+1th round update; According to the search state of the particle at the previous moment To minimize the fitness function f(X k ) as the goal, and find the local optimal solution at the next moment and the global optimal solution The specific expressions are as follows: Next, the particle's flight speed and position are iteratively updated using the currently explored optimal solution space until the iteration condition is no longer met, that is, when the maximum number of iterations N is reached, the optimal calculation migration strategy is output. In particular, a penalty amplification coefficient τ is introduced to make the penalty factor δ show a linear growth trend, so as to achieve the randomness of the early migration decision exploration and ensure the accuracy and stability of the later resource allocation strategy output. It is specifically expressed as: δ n+1 =τδ n , where δ n+1 Expressed as the penalty factor at the n+1th iteration, δ n It is expressed as the penalty factor at the nth iteration.

Images