A collaborative perception and resource allocation method based on edge computing

By adopting a collaborative perception and resource allocation method based on edge computing in industrial environments, the information transmission and resource allocation between sensors and edge estimators are optimized, the transmission delay and computing burden are solved, the state estimation performance is improved and the state estimation is guaranteed. System reliability and efficiency.

CN115065992BActive Publication Date: 2024-11-19DALIAN MARITIME UNIVERSITY
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202210518992.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-12
Publication Date
2024-11-19
Estimated Expiration
2042-05-12

AI Technical Summary

Technical Problem

In the prior art, in the process of transmitting sensor data to the remote control center in industrial environment monitoring, there is a transmission delay and a large calculation burden, resulting in a degradation of state estimation performance. A single control center is prone to failure to cause system paralysis and cannot effectively tolerate wireless channels. instability and random delay.

Method used

Adopting a collaborative perception and resource allocation method based on edge computing, by analyzing the impact of random transmission delay on state estimation error, a two-stage information transmission scheme and an improved marine predation algorithm are designed to optimize the correlation relationship between sensors and edge estimators. Subcarrier allocation and computing resource allocation achieve the minimization of fusion estimation errors.

Benefits of technology

It significantly reduces the computing burden of the remote control center, reduces transmission energy consumption and delay, improves the state estimation performance of the industrial environment, and ensures high reliability and low latency transmission of information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115065992B_ABST
    Figure CN115065992B_ABST
Patent Text Reader

Abstract

The present invention provides a collaborative perception and resource allocation method based on edge computing, which belongs to the field of wireless communication technology. The method includes the following steps: based on the state estimation method of edge computing, analyzing the influence of random transmission delay on state estimation error; designing a two-stage information transmission scheme based on edge collaboration, collaboratively optimizing the association between sensors and edge estimators, subcarrier allocation, and sensor transmission power, to minimize fusion estimation errors; based on an improved marine predation algorithm, determining the association between sensors and edge estimators, and realizing subcarrier dynamic allocation and computing resource allocation. The present invention effectively determines the association between sensors and edge estimators, realizes dynamic subcarrier allocation, can improve state perception accuracy, effectively solves the resource allocation problem in collaborative perception, and realizes efficient monitoring of complex industrial environments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular to a collaborative sensing and resource allocation method based on edge computing. Background Art

[0002] With the rapid development of wireless communication technology, wireless sensor networks are also showing new vitality in industrial Internet of Things systems, such as intelligent manufacturing, intelligent logistics, equipment fault detection, and production site monitoring. It has gradually become a trend to establish real-time connections between industrial field equipment using multi-sensor networks. A large number of sensors means that a large amount of data will be processed. If the information is not transmitted in time and accurately, it will cause serious losses to the factory and may even threaten human life and property safety. In previous industrial environment monitoring, wired transmission solutions or wireless transmission solutions are often used to transmit data to a remote data processing center. However, due to the harsh on-site environment and complex equipment, wiring is difficult and costly, making it difficult to deploy wired communication systems; in addition, all data is transmitted to the remote control center during the data transmission process, which brings a heavy communication and computing burden to the network, and ultimately makes it difficult to guarantee the state estimation performance of the remote control center.

[0003] In addition, due to the instability of the wireless channel, the perceived information cannot reach the remote control center on time. Then, under this condition, the remote control center must estimate the state of the next moment based on the previous prediction value or the measurement value of the sensor at the previous moment, but this may introduce additional estimation errors and affect the system estimation performance. Once the random delay exceeds a certain threshold, the state estimation performance will drop sharply. Therefore, it is crucial to ensure the transmission reliability between the sensor and the edge computing node. In addition, since the sensor's own computing power, communication capability and energy are limited, data processing at the sensor is not realistic. Therefore, in previous designs, all sensor perception information is transmitted to the remote control center. This not only brings a heavy computational burden to the control center, but also causes serious computational delays to deteriorate the state estimation performance.

[0004] In summary, the problems with the existing technology are: most of the current state monitoring designs use centralized state estimation, and all sensor observation data are transmitted to the remote control center. This will bring a heavy computing and communication burden to the network, resulting in increased delays; secondly, the remote center is generally placed far away from the site, which increases the transmission energy consumption; finally, under the premise of only one computing center, once the center fails, the entire system will be paralyzed. The current state perception algorithm passively tolerates the adverse effects of random delays on the transmission reliability and delay of industrial environment perception, which limits the improvement of industrial environment perception performance. Summary of the invention

[0005] To this end, the present invention proposes a collaborative perception and resource allocation method based on edge computing, comprising the following steps:

[0006] S1. State estimation method based on edge computing, analyzing the impact of random transmission delay on state estimation error;

[0007] S2: Design a two-stage information transmission scheme based on edge collaboration to collaboratively optimize the association between sensors and edge estimators, subcarrier allocation, and sensor transmit power to minimize the fusion estimation error;

[0008] S3. Based on the improved ocean predation algorithm, the correlation between the sensor and the edge estimator is determined to achieve dynamic subcarrier allocation and computing resource allocation.

[0009] Furthermore, the method is applicable to a network scenario in which the network scenario is an uplink perception information transmission network.

[0010] Furthermore, the state estimation method based on edge computing analyzes the influence of random transmission delay on state estimation error, and includes the following steps:

[0011] S101. Considering that the sensor perception information reaches the edge estimator with a delay, a random variable d is introduced. i (t) represents the random transmission delay experienced by the perception information of the i-th sensor, that is:

[0012] d i (t) = {d i (t|t)=0,d i (t+1|t)=1,…,d i (t+d i |t)=d i} (1)

[0013] in, Indicates the maximum delay allowed by the system; Δ = 10ms indicates the minimum delay allowed by the edge estimator within an estimation step;

[0014] and denote the transmission delay between the ith sensor and the mth edge estimator and the computational delay of the ith sensor on the mth edge estimator respectively;

[0015] S102, due to random delays, the sensor's perception information arrives at the edge estimator at different time steps, introducing a binary indicator variable To indicate which specific time step information is used by the edge estimator, the formula is as follows:

[0016]

[0017] in:

[0018] S103, the sensor perception information with random delay received by the edge estimator is expressed as:

[0019]

[0020] Where: y i (tj) represents the measurement information of the i-th sensor; x(tj) represents the actual state of the system; C i represents the measurement noise of the i-th sensor; j is the delay information of a specific time selected by the edge estimator; v i represents Gaussian white noise;

[0021] S104, based on the sensor perception information with random delay received by the edge estimator, the observation equation when the edge estimator performs local estimation can be expressed as:

[0022] y(t)=Cx(t)+v(t) (4)

[0023] in:

[0024]

[0025]

[0026] S105, when the edge estimator performs local estimation based on the received perception information, the recursive update step of its estimation value under random delay can be expressed as:

[0027]

[0028] P m (t|t-1) = AP m (t-1|t-1)A T +Q w

[0029]

[0030] P m (t|t)=(IK m (t)C)P m (t|t-1)(IK m (t)C) T +K m (t)χK m T (t)

[0031] Among them, K m (t) = P m (t|t-1)C T (CP m (t|t-1)C T +χ) represents the Kalman gain; χ represents the delay v i The estimated covariance matrix of w represents the variance of Gaussian white noise;

[0032] S106, after each edge estimator obtains a local estimation value, the local estimation value is transmitted between the edge estimators, and the fusion estimation at the edge estimator is expressed as:

[0033]

[0034] in, represents the fusion estimation weight;

[0035] I0=[I b ,…,I b ] T ;

[0036] The error variance matrix of the fused estimate at the edge estimator can be expressed as.

[0037]

[0038] Furthermore, the design is based on a two-stage information transmission scheme of edge collaboration, which collaboratively optimizes the association between the sensor and the edge estimator, the subcarrier allocation, and the sensor transmission power to minimize the fusion estimation error; and includes the following steps:

[0039] S201, dividing the transmission process between the multi-sensor and the edge estimator into two stages: in the first stage, each sensor provides raw measurement to the edge estimator associated with it, and in the second stage, each edge estimator performs local estimation based on the received raw measurement value, and then forwards the local estimation value to other edge estimators for distributed fusion estimation;

[0040] S202, determining the transmission delay in the first stage. The transmission delay between the i-th sensor and the m-th edge estimator is obtained by the following formula:

[0041]

[0042] Among them, Q i The size of the information sensed by each sensor; Represents the data transmission rate between the i-th sensor and the m-th edge estimator; binary variable indicates whether subcarrier k is allocated to the i-th sensor to pass the sensing information to the m-th edge estimator; p i,k is the transmission power of the ith sensor on the kth subcarrier, is the channel gain of the kth subcarrier when it is transmitted between the ith sensor and the mth edge estimator, σ 2 is the power of additive white Gaussian noise, and the energy consumption generated in the first stage is expressed as

[0043] S203, determining the computational delay in the second stage, the computational delay generated by the edge estimator processing the corresponding task is:

[0044]

[0045] Among them, D i The total number of CPU cycles required to complete the task; f i,m is the computing resource allocated by the mth edge estimator to process the perception information of the ith sensor. The computing energy consumption of the second stage is expressed as in Indicates the energy consumption coefficient of each CPU cycle;

[0046] S204. Due to the limitation of the computing capability of the edge estimator, the mth edge estimator needs to consider the following constraints when allocating computing resources to the sensor:

[0047]

[0048] where F m represents the maximum computing power of the mth edge estimator;

[0049] S205, determining the transmission delay in the second stage, the local estimated transmission delay between the mth edge estimator and the m'th edge estimator can be obtained by the following formula:

[0050]

[0051] in, Represents the data transmission rate between the mth edge estimator and the m'th edge estimator; binary variable Indicates whether to assign subcarrier k to the mth edge estimator to pass the local estimation information to the m'th edge estimator; p m is the transmission power of edge estimator m; is the channel gain of the kth subcarrier when it is transmitted between the mth edge estimator and the m'th edge estimator; R i It is the ratio of the task output information size to the task input information size;

[0052] S206, determining the association relationship between the sensor and the edge estimator, optimizing the subcarrier allocation and the sensor transmit power during the two-stage transmission process, and modeling it into a constrained optimization problem of minimizing the variance of the fusion estimation error, as shown below:

[0053]

[0054] Among them, C1 represents the bandwidth constraint for the transmission of perception information from the sensor to the edge estimator; C2 represents the bandwidth constraint for the transmission of local estimation information between edge estimators; C3 represents the transmission power constraint of each sensor; C4 represents the computing resource constraint of each edge estimator; C5 represents the maximum tolerable delay of the perception information of the i-th sensor; C6 represents the maximum tolerable delay for the exchange of local estimation information between edge estimators; C7 and C8 represent that each subcarrier can be occupied by at most one sensor / edge estimator. Solving this mixed integer nonlinear programming problem can obtain the association relationship between sensors and edge estimators, two-stage subcarrier allocation, and sensor transmit power allocation with the minimum variance of the fusion estimation error;

[0055] S207, according to the different contents of information transmitted in the two stages, the mixed integer nonlinear programming problem is divided into two sub-problems for solution: In the first stage, the sub-problem of sensor perception information transmission can be expressed as:

[0056]

[0057] Furthermore, given the transmission power, the subcarrier allocation and sensor-edge estimator association problem is expressed as:

[0058]

[0059] stC1,C4,C5,C7

[0060] The transmission power optimization under a given association relationship can be determined by solving the following problem:

[0061]

[0062] stC3,C5

[0063] S208. In the second stage, the local estimation transmission sub-problem is expressed as.

[0064]

[0065] Furthermore, the improved marine predation algorithm is used to determine the association relationship between the sensor and the edge estimator, and to achieve dynamic subcarrier allocation and computing resource allocation, which specifically includes the following steps:

[0066] S301, initializing the marine predation algorithm using a normalization factor and a reverse learning factor;

[0067] S302, running the improved ocean predation algorithm for the first time, obtaining a preference list of the edge estimator corresponding to each sensor according to the predator matrix generated by the improved ocean predation algorithm, wherein each row of the predator matrix represents a different preference list, and on this basis, calculating the computing power allocated to different sensors by the edge estimator;

[0068] S303, calculating the computing delay between the current sensor and the edge estimator according to the computing power allocated by the edge estimator;

[0069] S304, run the improved ocean predator algorithm for the second time, and obtain the preference list of the corresponding subcarriers under the sensor-edge estimator association relationship according to the predator matrix generated by the improved ocean predator algorithm. , where each row of the predator matrix represents a different preference list, on this basis, perform power allocation, and obtain the data transmission rate at the sensor and the edge estimator;

[0070] S305, calculating the transmission delay between the sensor and the edge estimator according to the data transmission rate at the sensor and the edge estimator;

[0071] S306, the edge estimator calculates the local estimation error variance at this time according to the obtained delayed perception information;

[0072] S307, allocating the subcarrier with the largest k-th channel gain to the edge estimator for transmitting local estimation values ​​between edge estimators, and calculating the current fusion estimation error variance;

[0073] S308. Repeat steps S301-S307, and retain the allocation scheme that minimizes the fusion estimation error in two adjacent iterations until the number of algorithm iterations is met.

[0074] A collaborative sensing and resource allocation device based on edge computing, comprising:

[0075] Analysis module: used for edge computing-based state estimation method to analyze the impact of random transmission delay on state estimation error;

[0076] Optimization module: used to design a two-stage information transmission scheme based on edge collaboration, to collaboratively optimize the relationship between sensors and edge estimators, subcarrier allocation, and sensor transmit power, to minimize fusion estimation errors;

[0077] Determination module: It is used to determine the association between the sensor and the edge estimator based on the improved ocean predation algorithm, and realize the dynamic allocation of subcarriers and computing resources.

[0078] The present invention provides a collaborative perception and resource allocation method based on edge computing, studies the state estimation method based on edge computing, and analyzes the influence of random transmission delay on the state estimation error; designs a two-stage information transmission scheme based on edge collaboration, and collaboratively optimizes the association between sensors and edge estimators, subcarrier allocation, and sensor transmission power to minimize the fusion estimation error; designs an improved marine predation algorithm to effectively determine the association between sensors and edge estimators, and realizes dynamic subcarrier allocation and computing resource allocation;

[0079] This method mainly targets the requirements of the system for state estimation performance during data transmission, and models the problem as minimizing the fusion estimation error of each estimation time step. Through the improved ocean predation algorithm optimization, the sensor is associated with the edge estimator that best suits it, matched to the channel with the best channel conditions, and the optimal solution to the problem is found. The original problem is decomposed into two sub-problems according to the different data transmission contents. The first sub-problem is generated during the data transmission between the sensor and the edge estimator, and the second sub-problem is generated during the data transmission between the edge estimators. The state estimation performance of industrial temperature monitoring is further improved by solving the two sub-problems step by step.

[0080] The present invention improves the performance of the marine predation algorithm by introducing a normalization factor and a reverse learning factor, thereby effectively solving the problem of sensor and edge estimator association and subcarrier allocation.

[0081] The present invention proposes a hierarchical network architecture for multi-sensor collaborative perception in an industrial network environment. The sensor is only responsible for collecting and sending data, and sends the collected data to multiple edge estimators at the edge of the network for data processing. In this mode, the computing burden of the remote control center can be significantly reduced, and unnecessary computing delays can be reduced; in addition, the edge estimator is generally placed not far from the industrial site, which can reduce the transmission energy consumption and transmission delay of the sensor. Under the premise of ensuring the state estimation performance of the edge estimator, the information is transmitted with high reliability and low latency.

[0082] Based on the above reasons, the present invention can be widely promoted in fields such as wireless communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] 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 some examples of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0084] Figure 1A flow chart of a collaborative perception and resource allocation method based on edge computing provided in an embodiment of the present invention;

[0085] Figure 2 A network scenario diagram used in an embodiment of the present invention;

[0086] Figure 3 This is a flow chart of an improved algorithm based on a marine predation algorithm according to an embodiment of the present invention;

[0087] Figure 4 A comparison diagram of local estimation and fusion estimation of an edge estimator according to an embodiment of the present invention;

[0088] Figure 5 This is a comparison chart of fusion estimation performance under different algorithms of an embodiment of the present invention;

[0089] Figure 6 This is a comparison diagram of average delays of different algorithms according to an embodiment of the present invention;

[0090] Figure 7 This is a comparison diagram of average energy consumption under different algorithms of an embodiment of the present invention;

[0091] Figure 8 is a comparison diagram of estimation performance under different numbers of edge estimators according to an embodiment of the present invention;

[0092] Fig. 9 This is a comparison diagram of average delays under different numbers of subcarriers according to an embodiment of the present invention. DETAILED DESCRIPTION

[0093] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 should fall within the scope of protection of the present invention.

[0094] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0095] like Figure 1 As shown, the present invention provides a collaborative perception and resource allocation method based on edge computing, comprising the following steps:

[0096] S1. State estimation method based on edge computing, analyzing the impact of random transmission delay on state estimation error;

[0097] S2. Design a two-stage information transmission scheme based on edge collaboration to collaboratively optimize the relationship between sensors and edge estimators, subcarrier allocation, and sensor transmit power to minimize fusion estimation errors.

[0098] S3. Based on the improved ocean predation algorithm, the correlation between the sensor and the edge estimator is effectively determined to realize the dynamic allocation of subcarriers and computing resources.

[0099] The application principle of the present invention is further described below in conjunction with the accompanying drawings.

[0100] like Figure 2 As shown, the network scenario applicable to the present invention is the uplink perception information transmission network. In the specified industrial Internet of Things application scenario, multiple sensors are deployed to achieve state estimation through collaborative perception. The hierarchical network structure composed of N sensors and M edge estimators is shown in Figure 2As shown in the figure. In this scenario, multiple sensor nodes are deployed in the industrial site to collect information such as temperature and humidity of the industrial site in a collaborative manner, and transmit the collected sensor perception information to the edge estimator associated with it through a wireless channel, so as to realize the real-time state estimation of the industrial site. In the hierarchical network structure, the entire transmission process can be divided into two stages. In the first stage, each sensor provides the original measurement value to the edge estimator associated with it. In the second stage, each edge estimator performs local estimation based on the received original measurement value, and then forwards the local estimation value to other edge estimators for distributed fusion estimation. Assume that the channel gain obeys the Rayleigh fading model. Divide the total bandwidth into K subcarriers, and the bandwidth of each subcarrier is ΔB. Let N = {1,2,…,N}, M = {1,2,…,M}, K = {1,2,…,K} be the set of sensors, edge estimators and subcarriers respectively.

[0101] The temperature state of the industrial process is the estimation object of the present invention. The temperature state is expressed as x(t+1)=Ax(t)+w(t). This state is sensed by multiple sensors and is expressed as y i (t) = C i x(t)+v i (t). represents the system state at time step t, and b represents the dimension of the state vector;

[0102] represents the measurement value of the i-th sensor at time step t, k i Represents the measurement vector y i The dimension of (t); Represents the state transition matrix of the system; represents the measurement matrix of the i-th sensor. and It has a mean of zero and a variance of Q w and Q vi Gaussian white noise, and w(t) and v i (t) are independent of each other,

[0103] In specific implementation, as a preferred embodiment of the present invention, the state estimation method based on edge computing analyzes the influence of random transmission delay on state estimation error, including the following steps:

[0104] S101. Considering that the sensor perception information reaches the edge estimator with a delay, a random variable d is introduced. i (t) represents the random transmission delay experienced by the perception information of the i-th sensor, that is:

[0105] d i (t) = {d i (t|t)=0,di (t+1|t)=1,…,d i (t+d i |t)=d i} (1)

[0106] in, Indicates the maximum delay allowed by the system; Δ = 10ms indicates the minimum delay allowed by the edge estimator within an estimation step;

[0107] and denote the transmission delay between the ith sensor and the mth edge estimator and the computational delay of the ith sensor on the mth edge estimator (described in detail in S2 ), respectively;

[0108] S102, due to random delays, the sensor's perception information arrives at the edge estimator at different time steps, introducing a binary indicator variable To indicate which specific time step information is used by the edge estimator, the formula is as follows:

[0109]

[0110] in:

[0111] S103, the sensor perception information with random delay received by the edge estimator is expressed as:

[0112]

[0113] Where: y i (tj) represents the measurement information of the i-th sensor; x(tj) represents the actual state of the system; C i represents the measurement noise of the i-th sensor; j is the delay information of a specific time selected by the edge estimator; v i represents Gaussian white noise;

[0114] S104, based on the sensor perception information with random delay received by the edge estimator, the observation equation when the edge estimator performs local estimation can be expressed as:

[0115] y(t)=Cx(t)+v(t) (4)

[0116] in:

[0117]

[0118]

[0119] S105, when the edge estimator performs local estimation based on the received perception information, the recursive update step of its estimation value under random delay can be expressed as:

[0120]

[0121] P m (t|t-1) = AP m (t-1|t-1)A T +Q w

[0122]

[0123] P m (t|t)=(IK m (t)C)P m (t|t-1)(IK m (t)C) T +K m (t)χK m T (t)

[0124] Among them, K m (t) = P m (t|t-1)C T (CP m (t|t-1)C T +χ) represents the Kalman gain; χ represents the delay v i The estimated covariance matrix of w represents the variance of Gaussian white noise;

[0125] S106, after each edge estimator obtains a local estimation value, the local estimation value is transmitted between edge estimators for the accuracy of estimation. Then the fusion estimation at the edge estimator is expressed as:

[0126]

[0127] in, represents the fusion estimation weight;

[0128] I0=[I b ,…,I b ] T ;

[0129] The error variance matrix of the fused estimate at the edge estimator can be expressed as.

[0130]

[0131] From the above analysis, we can see that P m′(t) is an implicit expression of random delay. The relationship between the sensor and the edge estimator, subcarrier allocation, and computing resource allocation will all affect the delay. Therefore, the joint optimization of the above factors is studied in S2 to reduce the delay as much as possible, thereby reducing the impact of random delay on state estimation.

[0132] In specific implementation, as a preferred embodiment of the present invention, the design is based on a two-stage information transmission scheme of edge collaboration, which collaboratively optimizes the association between the sensor and the edge estimator, the subcarrier allocation, and the sensor transmit power to minimize the fusion estimation error; specifically, it includes the following steps:

[0133] S201, the transmission process between multiple sensors and edge estimators is divided into two stages. In the first stage, each sensor provides raw measurements to the edge estimator associated with it, and in the second stage, each edge estimator performs local estimation based on the received raw measurements, and then forwards the local estimation values ​​to other edge estimators for distributed fusion estimation;

[0134] S202, determine the transmission delay in the first stage. The transmission delay between the i-th sensor and the m-th edge estimator is obtained by the following formula:

[0135]

[0136] Among them, Q i The size of the information sensed by each sensor; Represents the data transmission rate between the i-th sensor and the m-th edge estimator; binary variable indicates whether subcarrier k is allocated to the i-th sensor to pass the sensing information to the m-th edge estimator; p i,k is the transmission power of the ith sensor on the kth subcarrier, is the channel gain of the kth subcarrier when it is transmitted between the ith sensor and the mth edge estimator, σ 2 is the power of additive white Gaussian noise. In addition, the energy consumption generated in the first stage is expressed as

[0137] S203, determine the computational delay in the second stage. The computational delay generated by the edge estimator processing the corresponding task is:

[0138]

[0139] Among them, D i The total number of CPU cycles required to complete the task; f i,m is the computing resource allocated by the mth edge estimator to process the perception information of the ith sensor. The computing energy consumption of the second stage is expressed as in Indicates the energy consumption coefficient of each CPU cycle;

[0140] S204. Due to the limitation of the computing capability of the edge estimator, the mth edge estimator needs to consider the following constraints when allocating computing resources to the sensor:

[0141]

[0142] where F m represents the maximum computing power of the mth edge estimator;

[0143] S205, determine the transmission delay in the second stage. The locally estimated transmission delay between the mth edge estimator and the m'th edge estimator can be obtained by the following formula:

[0144]

[0145] in, Represents the data transmission rate between the mth edge estimator and the m'th edge estimator; binary variable Indicates whether to assign subcarrier k to the mth edge estimator to pass the local estimation information to the m'th edge estimator; p m is the transmission power of edge estimator m; is the channel gain of the kth subcarrier when it is transmitted between the mth edge estimator and the m'th edge estimator; R i It is the ratio of the task output information size to the task input information size;

[0146] S206, optimizing the association relationship between the sensor and the edge estimator, the subcarrier allocation during the two-stage transmission process, and the sensor transmission power to reduce the delay as much as possible, thereby reducing the fusion estimation error variance. This problem is modeled as a constrained optimization problem for minimizing the fusion estimation error variance, as shown below:

[0147]

[0148] Among them, C1 represents the bandwidth constraint of the sensor transmitting perception information to the edge estimator; C2 represents the bandwidth constraint of the transmission of local estimation information between edge estimators; C3 represents the transmission power constraint of each sensor; C4 represents the computing resource constraint of each edge estimator; C5 represents the maximum tolerable delay of the i-th sensor perception information; C6 represents the maximum tolerable delay of exchanging local estimation information between edge estimators; C7 and C8 represent that each subcarrier can be occupied by at most one sensor / edge estimator. Solving this mixed integer nonlinear programming problem can obtain the association relationship between sensors and edge estimators, two-stage subcarrier allocation, and sensor transmission power allocation with the minimum variance of fusion estimation error.

[0149] S207: To facilitate solving the problem in S206, the problem is divided into two sub-problems according to the different contents of information transmitted in the two stages. In the first stage, the sub-problem of sensor perception information transmission can be expressed as:

[0150]

[0151] Furthermore, given the transmission power, the subcarrier allocation and sensor-edge estimator association problem is expressed as:

[0152]

[0153] stC1,C4,C5,C7

[0154] The transmission power optimization under a given association relationship can be determined by solving the following problem:

[0155]

[0156] stC3,C5

[0157] S208, in the second stage, the local estimation transmission sub-problem is expressed as:

[0158]

[0159] Figure 3 This is a flow chart of an improved algorithm based on a marine predation algorithm according to an embodiment of the present invention;

[0160] The improved marine predation algorithm is used to determine the association relationship between the sensor and the edge estimator, and to achieve dynamic subcarrier allocation and computing resource allocation, which specifically includes the following steps:

[0161] S301, initializing the marine predation algorithm using a normalization factor and a reverse learning factor;

[0162] S302, run the improved ocean predation algorithm for the first time, and obtain the preference list of the edge estimator corresponding to each sensor according to the predator matrix generated by the improved ocean predation algorithm. Each row of the predator matrix represents a different preference list, and on this basis, calculate the computing power allocated to different sensors by the edge estimator;

[0163] S303, calculating the computing delay between the current sensor and the edge estimator according to the computing power allocated by the edge estimator;

[0164] S304, run the improved ocean predator algorithm for the second time, and obtain the preference list of the corresponding subcarriers under the sensor-edge estimator association relationship according to the predator matrix generated by the improved ocean predator algorithm. Each row of the predator matrix represents a different preference list, and on this basis, power allocation is performed to obtain the data transmission rate at the sensor and the edge estimator;

[0165] S305, calculating the transmission delay between the sensor and the edge estimator according to the data transmission rate at the sensor and the edge estimator;

[0166] S306, the edge estimator calculates the local estimation error variance at this time according to the obtained delayed perception information;

[0167] S307, allocating the subcarrier with the largest k-th channel gain to the edge estimator for transmitting local estimation values ​​between edge estimators, and calculating the current fusion estimation error variance;

[0168] S308. Repeat steps S301-S307, and retain the allocation scheme that minimizes the fusion estimation error in two adjacent iterations until the number of algorithm iterations is met.

[0169] A collaborative sensing and resource allocation device based on edge computing, comprising:

[0170] Analysis module: used for edge computing-based state estimation method to analyze the impact of random transmission delay on state estimation error;

[0171] Optimization module: used to design a two-stage information transmission scheme based on edge collaboration, to collaboratively optimize the relationship between sensors and edge estimators, subcarrier allocation, and sensor transmit power, to minimize fusion estimation errors;

[0172] Determination module: It is used to determine the association between the sensor and the edge estimator based on the improved ocean predation algorithm, and realize the dynamic allocation of subcarriers and computing resources.

[0173] The application effect of the present invention is described in detail below in conjunction with simulation:

[0174] 1. Simulation conditions

[0175] In the simulation scenario, the monitoring range in the industrial environment is [0,50] m *[0,10] m, 3 edge estimators and 6 sensors are randomly arranged within this range. The total bandwidth of the wireless channel is 10MHz, the number of subcarriers is 16, the maximum transmission power of the sensor is 10dBm, and the maximum transmission power of the edge estimator is 20dBm. The power spectral density of Gaussian white noise in the industrial environment is -87dBm / Hz, the size of the perception data of each sensor is between 200-500Kbits, the total computing power of the edge estimator is between 2-4GHz, the number of CPU cycles required for the sensor task is between 300-600MCycles, the maximum number of iterations is 1500, and the number of predators is 25.

[0176] 2. Simulation content and results analysis

[0177] The performance of the proposed algorithm is demonstrated by comparing it with other transmission methods.

[0178] Comparative Algorithm 1—Particle Swarm Optimization: The association between the sensor and the edge estimator and the subcarrier allocation problem are considered as a particle in the search space, and the solution is iteratively solved in the solution space by changing the speed and direction of the particle.

[0179] Comparison Algorithm 2 - Random Algorithm: Randomly determine the association between the sensor and the edge estimator, and randomly assign subcarriers to the sensor and the edge estimator. Based on this, calculate the fusion estimation performance at the edge estimator under the random strategy.

[0180] Simulation 1: Analysis of the comparison between the local estimation error and the fusion estimation error at the edge node.

[0181] from Figure 4 It can be seen that since the edge estimator only receives the perception information of a part of the sensors in the local estimation stage, and the received information is incomplete and delayed, the local estimation error is greater than the fusion estimation error in the final fusion stage. In the final fusion estimation stage, the edge estimator indirectly receives the perception information of all sensors, so the fusion estimation errors of all edge estimators eventually converge to a consistent value. In particular, we can see that in the initial stage, the estimation errors of different edge estimators are different. This is because each edge estimator receives perception information from different sensors with different delays.

[0182] Simulation 2: Analyze the performance comparison under different algorithms.

[0183] from Figure 5-7 It can be seen that under the premise that all simulation parameters are consistent, the improved marine predation algorithm performs better than the other two algorithms. Figure 5It can be seen that the system fusion estimation performance under the optimization of the ocean predation algorithm is significantly better than the other two algorithms. This is because the fusion estimation error is highly dependent on the delay, and in the ocean predation algorithm, the average delay of all sensors reaching each edge estimator (such as Figure 6 ) is the smallest, so the best system performance is obtained under the improved ocean predation algorithm. Figure 7 The average transmission energy consumption of the sensor reaching each edge estimator under different algorithm scheduling is shown. From this, we can see that the average transmission energy consumption under the Particle Swarm Optimization Algorithm and the Ocean Predator Algorithm scheduling is basically the same, while the energy consumption under the random allocation algorithm is the largest. This is because compared with the other two algorithms, the random allocation scheme does not optimize the association between the sensor and the edge estimator and the subcarrier allocation, resulting in higher energy consumption.

[0184] from Figure 6 It can be seen that on some edge estimators, the average delay of the particle swarm algorithm is higher than that of the random algorithm. This is because there are fewer sensors associated with the edge estimator in the random algorithm compared to the particle swarm algorithm, so the estimation error of the random edge estimator is larger than that of the particle swarm algorithm. It is precisely because fewer sensor-edge estimator associations are established under the random algorithm that the sensors obtain more communication and computing resources under the random algorithm scheduling. Therefore, the delay of the particle swarm algorithm is larger than that of the random algorithm.

[0185] from Figure 8 It can be seen that as the number of edge estimators increases, the estimation performance of the system becomes better. This is because in the case of a single edge estimator, the system cannot guarantee timely processing of the sensor's perception information, which makes the processing delay in the fusion estimation stage greater. With multiple edge estimators, the system can pre-process the perception information in a timely manner, thereby ensuring the estimation performance. Fig. 9 It can be seen that as the number of subcarriers increases, the latency increases slightly, but the latency of the improved ocean predation algorithm is always lower than that of the other two algorithms. This is reasonable because the number of subcarriers affects the number of candidate associations between the sensor and the edge estimator. With the same number of subcarriers, the particle swarm algorithm and the random algorithm establish more candidate associations between the sensor and the edge estimator than the ocean predation algorithm, resulting in higher latency.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative perception and resource allocation method based on edge computing, It is characterized in that The steps include: S1. State estimation method based on edge computing, analyzing the impact of random transmission delay on state estimation error; S2: Design a two-stage information transmission scheme based on edge collaboration to collaboratively optimize the association between sensors and edge estimators, subcarrier allocation, and sensor transmit power to minimize the fusion estimation error; The design is based on a two-stage information transmission scheme of edge collaboration, which collaboratively optimizes the association between sensors and edge estimators, subcarrier allocation, and sensor transmit power to minimize the fusion estimation error; it includes the following steps: S201, dividing the transmission process between the multi-sensor and the edge estimator into two stages: in the first stage, each sensor provides raw measurement to the edge estimator associated with it, and in the second stage, each edge estimator performs local estimation based on the received raw measurement value, and then forwards the local estimation value to other edge estimators for distributed fusion estimation; S202, determining the transmission delay in the first stage. The transmission delay between the i-th sensor and the m-th edge estimator is obtained by the following formula: Among them, Q i The size of the information sensed by each sensor; Represents the data transmission rate between the i-th sensor and the m-th edge estimator; binary variable indicates whether subcarrier k is allocated to the i-th sensor to pass the sensing information to the m-th edge estimator; p i,k is the transmission power of the ith sensor on the kth subcarrier, is the channel gain of the kth subcarrier when it is transmitted between the ith sensor and the mth edge estimator, σ 2 is the power of additive white Gaussian noise, and the energy consumption generated in the first stage is expressed as S203, determining the computational delay in the second stage, the computational delay generated by the edge estimator processing the corresponding task is: Among them, D i The total number of CPU cycles required to complete the task; f i,m is the computing resource allocated by the mth edge estimator to process the perception information of the ith sensor. The computing energy consumption of the second stage is expressed as in Indicates the energy consumption coefficient of each CPU cycle; S204. Due to the limitation of the computing capability of the edge estimator, the mth edge estimator needs to consider the following constraints when allocating computing resources to the sensor: Among them, F m represents the maximum computing power of the mth edge estimator; S205, determining the transmission delay in the second stage, the local estimated transmission delay between the mth edge estimator and the m'th edge estimator can be obtained by the following formula: in, Represents the data transmission rate between the mth edge estimator and the m'th edge estimator; binary variable Indicates whether to assign subcarrier k to the mth edge estimator to pass the local estimation information to the m'th edge estimator; p m is the transmission power of edge estimator m; is the channel gain of the kth subcarrier when it is transmitted between the mth edge estimator and the m'th edge estimator; R i It is the ratio of the task output information size to the task input information size; S206, determining the association relationship between the sensor and the edge estimator, optimizing the subcarrier allocation and the sensor transmit power during the two-stage transmission process, and modeling it into a constrained optimization problem of minimizing the variance of the fusion estimation error, as shown below: Among them, C1 represents the bandwidth constraint for the transmission of perception information from the sensor to the edge estimator; C2 represents the bandwidth constraint for the transmission of local estimation information between edge estimators; C3 represents the transmission power constraint of each sensor; C4 represents the computing resource constraint of each edge estimator; C5 represents the maximum tolerable delay of the perception information of the i-th sensor; C6 represents the maximum tolerable delay for the exchange of local estimation information between edge estimators; C7 and C8 represent that each subcarrier can be occupied by at most one sensor / edge estimator. Solving this mixed integer nonlinear programming problem can obtain the association relationship between sensors and edge estimators, two-stage subcarrier allocation, and sensor transmit power allocation with the minimum variance of the fusion estimation error; S207, according to the different contents of information transmitted in the two stages, the mixed integer nonlinear programming problem is divided into two sub-problems for solution: In the first stage, the sub-problem of sensor perception information transmission can be expressed as: Furthermore, given the transmission power, the subcarrier allocation and sensor-edge estimator association problem is expressed as: stC1,C4,C5,C7 The transmission power optimization under a given association relationship can be determined by solving the following problem: stC3,C5 S208, in the second stage, the local estimation transmission sub-problem is expressed as: S3, based on the improved marine predation algorithm, the association between the sensor and the edge estimator is determined to achieve dynamic subcarrier allocation and computing resource allocation; The improved marine predation algorithm is used to determine the association relationship between the sensor and the edge estimator, and to achieve dynamic subcarrier allocation and computing resource allocation, which specifically includes the following steps: S301, initializing the marine predation algorithm using a normalization factor and a reverse learning factor; S302, running the improved ocean predation algorithm for the first time, obtaining a preference list of the edge estimator corresponding to each sensor according to the predator matrix generated by the improved ocean predation algorithm, wherein each row of the predator matrix represents a different preference list, and on this basis, calculating the computing power allocated to different sensors by the edge estimator; S303, calculating the computing delay between the current sensor and the edge estimator according to the computing power allocated by the edge estimator; S304, running the improved ocean predation algorithm for the second time, obtaining a preference list of corresponding subcarriers under the sensor-edge estimator association relationship according to the predator matrix generated by the improved ocean predation algorithm, wherein each row of the predator matrix represents a different preference list, and on this basis, performing power allocation to obtain the data transmission rate at the sensor and the edge estimator; S305, calculating the transmission delay between the sensor and the edge estimator according to the data transmission rate at the sensor and the edge estimator; S306, the edge estimator calculates the local estimation error variance at this time according to the obtained delayed perception information; S307, allocating the subcarrier with the largest k-th channel gain to the edge estimator for transmitting local estimation values ​​between edge estimators, and calculating the current fusion estimation error variance; S308. Repeat steps S301-S307, and retain the allocation scheme that minimizes the fusion estimation error in two adjacent iterations until the number of algorithm iterations is met.

2. According to the collaborative perception and resource allocation method based on edge computing according to claim 1, It is characterized in that The method is applicable to a network scenario in which the network is an uplink perception information transmission network.

3. According to the collaborative perception and resource allocation method based on edge computing according to claim 1, It is characterized in that The state estimation method based on edge computing analyzes the influence of random transmission delay on state estimation error, and includes the following steps: S101. Considering that the sensor perception information reaches the edge estimator with a delay, a random variable d is introduced. i (t) represents the random transmission delay experienced by the perception information of the i-th sensor, that is: d i (t)={d i (t|t)=0,d i (t+1|t)=1,…,d i (t+d i |t)=d i } (1) in, Indicates the maximum delay allowed by the system; Δ = 10ms indicates the minimum delay allowed by the edge estimator within an estimation step; and denote the transmission delay between the ith sensor and the mth edge estimator and the computational delay of the ith sensor on the mth edge estimator respectively; S102, due to random delays, the sensor's perception information arrives at the edge estimator at different time steps, introducing a binary indicator variable To indicate which specific time step information is used by the edge estimator, the formula is as follows: in: S103, the sensor perception information with random delay received by the edge estimator is expressed as: Where: y i (tj) represents the measurement information of the i-th sensor; x(tj) represents the actual state of the system; C i represents the measurement noise of the i-th sensor; j is the delay information of a specific time selected by the edge estimator; v i represents Gaussian white noise; S104, based on the sensor perception information with random delay received by the edge estimator, the observation equation when the edge estimator performs local estimation can be expressed as: y(t)=Cx(t)+v(t) (4) in: C=diag[C 1 T ,…,C N T ] T ; S105, when the edge estimator performs local estimation based on the received perception information, the recursive update step of its estimation value under random delay can be expressed as: P m (t|t-1)=AP m (t-1|t-1)A T +Q w P m (t|t)=(I-K m (t)C)P m (t|t-1)(I-K m (t)C) T +K m (t)χK m T (t) Among them, K m (t) = P m (t|t-1)C T (CP m (t|t-1)C T +χ) represents the Kalman gain; χ represents the delay v i The estimated covariance matrix of w represents the variance of Gaussian white noise; S106, after each edge estimator obtains a local estimation value, the local estimation value is transmitted between the edge estimators, and the fusion estimation at the edge estimator is expressed as: in, represents the fusion estimation weight; I 0 =[I b ,…,I b ] T ; The error variance matrix of the fused estimate at the edge estimator can be expressed as 。 4. A collaborative perception and resource allocation device based on edge computing, It is characterized in that include Analysis module: used for edge computing-based state estimation method to analyze the impact of random transmission delay on state estimation error; Optimization module: used to design a two-stage information transmission scheme based on edge collaboration, to collaboratively optimize the relationship between sensors and edge estimators, subcarrier allocation, and sensor transmit power, to minimize fusion estimation errors; The design is based on a two-stage information transmission scheme of edge collaboration, which collaboratively optimizes the association between sensors and edge estimators, subcarrier allocation, and sensor transmit power to minimize the fusion estimation error; it includes the following steps: S201, dividing the transmission process between the multi-sensor and the edge estimator into two stages: in the first stage, each sensor provides raw measurement to the edge estimator associated with it, and in the second stage, each edge estimator performs local estimation based on the received raw measurement value, and then forwards the local estimation value to other edge estimators for distributed fusion estimation; S202, determining the transmission delay in the first stage. The transmission delay between the i-th sensor and the m-th edge estimator is obtained by the following formula: Among them, Q i The size of the information sensed by each sensor; Represents the data transmission rate between the i-th sensor and the m-th edge estimator; binary variable indicates whether subcarrier k is allocated to the i-th sensor to pass the sensing information to the m-th edge estimator; p i,k is the transmission power of the ith sensor on the kth subcarrier, is the channel gain of the kth subcarrier when it is transmitted between the ith sensor and the mth edge estimator, σ 2 is the power of additive white Gaussian noise, and the energy consumption generated in the first stage is expressed as S203, determining the computational delay in the second stage, the computational delay generated by the edge estimator processing the corresponding task is: Among them, D i The total number of CPU cycles required to complete the task; f i,m is the computing resource allocated by the mth edge estimator to process the perception information of the ith sensor. The computing energy consumption of the second stage is expressed as in Indicates the energy consumption coefficient of each CPU cycle; S204. Due to the limitation of the computing capability of the edge estimator, the mth edge estimator needs to consider the following constraints when allocating computing resources to the sensor: Among them, F m represents the maximum computing power of the mth edge estimator; S205, determining the transmission delay in the second stage, the local estimated transmission delay between the mth edge estimator and the m'th edge estimator can be obtained by the following formula: in, Represents the data transmission rate between the mth edge estimator and the m'th edge estimator; binary variable Indicates whether to assign subcarrier k to the mth edge estimator to pass the local estimation information to the m'th edge estimator; p m is the transmission power of edge estimator m; is the channel gain of the kth subcarrier when it is transmitted between the mth edge estimator and the m'th edge estimator; R i It is the ratio of the task output information size to the task input information size; S206, determining the association relationship between the sensor and the edge estimator, optimizing the subcarrier allocation and the sensor transmit power during the two-stage transmission process, and modeling it into a constrained optimization problem of minimizing the variance of the fusion estimation error, as shown below: Among them, C1 represents the bandwidth constraint for the transmission of perception information from the sensor to the edge estimator; C2 represents the bandwidth constraint for the transmission of local estimation information between edge estimators; C3 represents the transmission power constraint of each sensor; C4 represents the computing resource constraint of each edge estimator; C5 represents the maximum tolerable delay of the perception information of the i-th sensor; C6 represents the maximum tolerable delay for the exchange of local estimation information between edge estimators; C7 and C8 represent that each subcarrier can be occupied by at most one sensor / edge estimator. Solving this mixed integer nonlinear programming problem can obtain the association relationship between sensors and edge estimators, two-stage subcarrier allocation, and sensor transmit power allocation with the minimum variance of the fusion estimation error; S207, according to the different contents of information transmitted in the two stages, the mixed integer nonlinear programming problem is divided into two sub-problems for solution: In the first stage, the sub-problem of sensor perception information transmission can be expressed as: Furthermore, given the transmission power, the subcarrier allocation and sensor-edge estimator association problem is expressed as: stC1,C4,C5,C7 The transmission power optimization under a given association relationship can be determined by solving the following problem: stC3,C5 S208, in the second stage, the local estimation transmission sub-problem is expressed as: Determination module: used to determine the association between the sensor and the edge estimator based on the improved marine predation algorithm, and realize the dynamic allocation of subcarriers and computing resources; The improved marine predation algorithm is used to determine the association relationship between the sensor and the edge estimator, and to achieve dynamic subcarrier allocation and computing resource allocation, which specifically includes the following steps: S301, initializing the marine predation algorithm using a normalization factor and a reverse learning factor; S302, running the improved ocean predation algorithm for the first time, obtaining a preference list of the edge estimator corresponding to each sensor according to the predator matrix generated by the improved ocean predation algorithm, wherein each row of the predator matrix represents a different preference list, and on this basis, calculating the computing power allocated to different sensors by the edge estimator; S303, calculating the computing delay between the current sensor and the edge estimator according to the computing power allocated by the edge estimator; S304, running the improved ocean predation algorithm for the second time, obtaining a preference list of corresponding subcarriers under the sensor-edge estimator association relationship according to the predator matrix generated by the improved ocean predation algorithm, wherein each row of the predator matrix represents a different preference list, and on this basis, performing power allocation to obtain the data transmission rate at the sensor and the edge estimator; S305, calculating the transmission delay between the sensor and the edge estimator according to the data transmission rate at the sensor and the edge estimator; S306, the edge estimator calculates the local estimation error variance at this time according to the obtained delayed perception information; S307, allocating the subcarrier with the largest k-th channel gain to the edge estimator for transmitting local estimation values ​​between edge estimators, and calculating the current fusion estimation error variance; S308. Repeat steps S301-S307, and retain the allocation scheme that minimizes the fusion estimation error in two adjacent iterations until the number of algorithm iterations is met.

Citation Information

Patent Citations

  • Unmanned ship cooperative transmission method for marine environment perception

    CN111586633A

  • Multi-user deep neural network model segmentation and resource allocation optimization method in edge computing scene

    CN112822701A

  • Communication and computing resource joint optimization method based on mobile edge computing

    CN113286329A

  • Energy efficient discontinuous mobile edge computing with quality of service guarantees

    EP3826368A1