Industrial Internet of Things resource offloading method based on Hurst and dual-population genetic algorithm

Abstract

The present invention provides an industrial Internet of Things resource offloading method based on Hurst and dual-population genetic algorithms, and relates to the field of resource scheduling technology. The method first constructs an industrial Internet of Things resource offloading system model including one D2D device and N user-type devices; secondly, the tasks generated by the N user-type devices are offloaded to local computing, D2D device computing, and MEC server computing, respectively, and an optimization goal P is set to minimize total energy consumption and latency; then, the historical computing pressure information of the one D2D device is normalized, and a D2D device candidate set is constructed based on the Hurst index H of the D2D device and the statistic S of the historical computing pressure of the D2D device obtained by the Cox-Stuart test method; finally, the D2D device candidate set is used to assist the MEC network server in task offloading, and a dual-population genetic algorithm is used for iteration to obtain the optimal offloading strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and in particular to an industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithms. Background Art

[0002] Considering that D2D (device-to-device) offloading can complement MEC (mobile edge computing) networks, related research has been extensively explored. To minimize task latency and terminal energy consumption in MEC networks, many researchers have proposed a joint optimization problem for task offloading and computational resource allocation in D2D-assisted MEC networks by establishing multi-objective constrained cost optimization models. Some studies have proposed a D2D computation task offloading scheme for federated learning. This scheme, which considers that different edge nodes exchange data samples via D2D communication, balances the processing power and task load of the nodes to minimize the overall latency of the federated learning model training process. Inspired by the challenges posed by unstable motion, selfish device behavior, and incomplete offloading information, some researchers have proposed a collaborative offloading framework that integrates migration cost and offloading willingness in D2D-assisted MEC networks. Furthermore, a learning-based collaborative task offloading algorithm has been developed. This algorithm enables IIoT (Industrial Internet of Things) devices to select the optimal edge node by observing and learning the system cost of candidate edge nodes, without requiring complete offloading information, to minimize system task latency and migration cost. Some researchers have noticed that unloading a single edge server may lead to idle and wasted resources of remote edge servers. To solve this problem, they proposed an edge cloud collaborative computing offloading strategy based on an improved hybrid particle swarm algorithm under a joint offloading scheme of cloud servers and multiple edge servers.

[0003] Previous research has not considered the availability of D2D devices when using them for task offloading. Some D2D devices are idle for extended periods or remain in a relatively stable, low-load state. In these cases, task offloading to these devices can achieve a higher quality of service. However, when offloading tasks to D2D devices with a heavy workload or a tendency to increase, these devices cannot provide reliable computing resources, resulting in significant latency and difficulty meeting the required quality of service. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the above-mentioned existing technologies and provide an industrial Internet of Things resource offloading method based on Hurst and dual-population genetic algorithms. The method analyzes the historical computing pressure of D2D devices (the percentage of computing resources used by the devices to the total computing resources at each time node in the past) and predicts the future trend of computing pressure, so as to select suitable D2D devices for task offloading.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: an industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithm, comprising the following steps:

[0006] Step 1: Build an industrial IoT resource offloading system model, including I D2D devices and N user-type devices;

[0007] Step 2: Split the tasks generated by N user type devices into three parts and offload them to local computing, D2D device computing, and MEC server computing respectively;

[0008] Step 3: Calculate the latency and energy consumption of the computation and offloading processes of the three parts after the split, and set the optimization goal P to minimize the total energy consumption and latency.

[0009] The optimization goal P is to minimize the total energy consumption and delay, which is expressed as minimizing the difference between the energy consumption and delay after overall offloading and the energy consumption and delay after only local offloading. The specific formula is as follows:

[0010]

[0011] Where, t n represents the total delay caused by the overall offloading of user type devices, E n Represents the total energy consumption generated by overall unloading of user-type devices, t n,x represents the delay when the task is only offloaded locally, E n,x represents the energy consumption when the task is only offloaded locally, α is the weight of the total delay, and β is the weight of the total energy consumption;

[0012] Step 4: Obtain historical calculated pressure information of I D2D devices in the industrial Internet of Things;

[0013] Step 5: Normalize all historical calculated pressure information of the D2D device;

[0014] Step 6: Construct a candidate set of D2D devices based on the Hurst index H of the D2D devices and the statistic S of the historical computing pressure of the D2D devices obtained by the Cox-Stuart test method;

[0015] Step 6-1: Calculate the Hurst index H of the D2D device based on the historical calculated pressure information of the D2D device;

[0016] When H = 0.5, it indicates that the historical calculated pressure performs a random walk without a fixed trend. At this time, the D2D device should be retained in the D2D device candidate set; when H < 0.5, it indicates that the calculated pressure has a reverse trend with respect to past changes and has anti-persistence, that is, the trend is more likely to reverse in the short term. At this time, the D2D device with a decreasing trend in historical calculated pressure should be removed from the D2D device candidate set; when H > 0.5, it indicates that the calculated pressure has a persistent trend with respect to past changes and has persistence, that is, the trend is more likely to continue in the short term. At this time, the D2D device with an increasing trend in historical calculated pressure should be removed from the D2D device candidate set;

[0017] Step 6-2: Use the Cox-Stuart test method to obtain the statistic S of the historical calculated pressure of the D2D device, and judge the change trend of the historical calculated pressure of the current D2D device according to the statistic S;

[0018] When 0.5 < S ≤ 1, it indicates that the historical calculated pressure of the device has a strong increasing trend. When -1 < S ≤ -0.5, it indicates that the historical calculated pressure of the device has a strong decreasing trend;

[0019] Step 6-3: Exclude the D2D devices that satisfy 0.5 < S ≤ 1, H > 0.5 and -1 < S ≤ -0.5, H < 0.5 from the candidate set according to the Hurst index H and the statistic S obtained by the Cox-Stuart test method;

[0020] Step 7: Use the D2D device candidate set to assist the MEC network server for task offloading, and use the double-population genetic algorithm for iteration to obtain the optimal offloading strategy;

[0021] When the double-population genetic algorithm is used for iteration to obtain the optimal offloading strategy, first generate the first-generation individuals in the solution space in a random generation manner; then calculate the fitness of the individuals in the population according to the domain optimization objective, and divide the population into two populations, the dominant population and the inferior population, according to the individual fitness. Set different crossover rates and mutation rates for the dominant population and the inferior population respectively, and perform independent crossover and mutation respectively; secondly, use the fast non-dominated sorting algorithm to screen out two populations for the next iteration according to the scale; finally, merge the two populations into the same population, and repeat the above operations until the maximum number of iterations is reached. At this time, select the individual with the highest fitness and perform decoding operation on it to obtain the optimal offloading strategy;

[0022] The crossover rate P of the dominant population c and the mutation rate p m are specifically expressed as:

[0023]

[0024] Where, P c max is the maximum value of the crossover rate, P c min is the minimum value of the crossover rate, is the maximum value of the mutation rate, is the minimum value of the mutation rate, g is the current number of iterations, g max is the maximum number of iterations.

[0025] The crossover rate of the inferior population is the maximum crossover rate of the superior population, expressed as P c max , the mutation rate of the inferior population is the maximum mutation rate of the dominant population, expressed as

[0026] The beneficial effects of adopting the above technical solution are: the resource offloading method in the industrial Internet of Things based on the Hurst index and the dual-population genetic algorithm provided by the present invention ensures that the D2D devices performing task offloading can provide high-quality services by selecting available D2D devices, and can significantly reduce computing latency and device energy consumption. Moreover, the present invention adopts a dual-population genetic algorithm for iteration, which can better explore the solution space and converge to the optimal solution more quickly than the traditional genetic algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A specific execution flow chart of the industrial Internet of Things resource offloading method based on Hurst and dual-population genetic algorithm provided in an embodiment of the present invention;

[0028] Figure 2 This is a flowchart of the execution of the dual-population genetic algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0030] In the embodiment of the present invention, the industrial Internet of Things resource unloading method based on Hurst and dual population genetic algorithm is as follows: Figure 1 As shown, the following steps are included:

[0031] Step 1: Build an industrial IoT resource offloading system model, including I D2D devices and N user-type devices;

[0032] In this embodiment, based on the device type, it is assumed that the total number of user-type devices is n = {1, 2, 3, 4, 5, ..., N}, and the number of D2D devices is i = {1, 2, 3, 4, 5, ..., I}. Assuming that each user-type device n generates a splittable computation-intensive task in each time period, the computation-intensive task generated by the user-type device is expressed as: T n ={D n ,C n ,L n}, where D n Indicates the data size of the current task, expressed in bits; C n Indicates the average number of CPU cycles required to complete the calculation of the current task unit size data; L n Indicates the maximum delay that the current task can tolerate.

[0033] Step 2: Split the tasks generated by N user type devices into three parts and offload them to local computing, D2D device computing, and MEC server computing respectively;

[0034] In this embodiment, after determining the unloading location, the task can be divided into three parts: D n,l , D n,i , D n,e , respectively representing offloading the task to the local execution part, offloading the task to the D2D device execution part, and offloading the task to the MEC server execution part. The specific representation is as follows: u n,l ={0,1} indicates whether to perform local uninstallation, 0 indicates not to perform local uninstallation, 1 indicates to perform local uninstallation; u n,e = {0,1} indicates whether to uninstall the MEC server, 0 indicates not to uninstall the MEC server, 1 indicates to uninstall the MEC server; u n,i ={0,1} Similarly, it indicates whether D2D offloading is performed, 0 indicates not performing D2D offloading, and 1 indicates performing D2D offloading.

[0035] Step 3: Calculate the latency and energy consumption of the task computation and offloading processes generated by the three types of user devices, and set the optimization goal to minimize the total energy consumption and latency.

[0036] In this embodiment, the CPU frequency allocated to local computing by the user type device is f n,l , then the delay caused by the local offloading part Expressed as:

[0037]

[0038] In this embodiment, the use of MEC servers for auxiliary calculations requires data upload, MEC server processing, and result return. Since the volume of the calculation results is usually small, the return delay can be ignored. The total delay caused by the MEC server-assisted calculation is approximately the sum of the upload delay and the calculation delay. n,e , expressed as:

[0039]

[0040]

[0041] Where, Indicates the delay in transmitting data to the MEC server. represents the delay caused by the calculation after the data is transmitted to the server, B n,e is the link bandwidth between the user type device and the MEC server, p n,e is the transmission power of user type devices to MEC servers, h n,e represents the channel gain, δ 2 represents the channel noise power. n,e The CPU frequency allocated to user-type devices by the MEC server.

[0042] In this embodiment, the delay caused by the user type device uploading the task to the D2D device Expressed as:

[0043]

[0044] Where B n,i P represents the link bandwidth between user type devices and D2D devices. n,i is the transmission power of the user type device to the D2D device, h n,i represents the channel gain, δ 2 Represents the channel noise power.

[0045] In this embodiment, the delay generated by the auxiliary calculation of the D2D device Expressed as:

[0046]

[0047] Where, f n,i Indicates the CPU frequency allocated by the D2D device to the user type device.

[0048] At this time, the total delay t generated by the D2D device assistance calculation n,i Expressed as:

[0049]

[0050] Therefore, the total delay t caused by the user type device processing task n It is expressed as the maximum delay of the local offloading part, the D2D offloading part, and the MEC server offloading part, which is expressed as follows:

[0051] t n =max{t n,l ,t n,e ,t n,i}

[0052] In this embodiment, for the task energy consumption, the total energy consumption E generated by the local offload calculation of the user type device, the D2D device offload calculation, and the MEC server offload calculation is n Expressed as:

[0053]

[0054] Where δ is the energy consumption coefficient of the CPU.

[0055] In this embodiment, only the delay t of local unloading is performed. n,x and energy consumption E n,x Expressed as:

[0056]

[0057] Where, f n Indicates the task processing frequency of the local CPU when no offloading is performed.

[0058] In this embodiment, the optimization goal P of minimizing the total energy consumption and delay is expressed as:

[0059]

[0060] Where α is the weight of total delay, and β is the weight of total energy consumption.

[0061] Step 4: Obtain historical calculated pressure information of I D2D devices in the industrial Internet of Things;

[0062] Step 5: Normalize all historical calculated pressure information of the D2D device. The specific method is as follows:

[0063]

[0064] Where x(t) represents the computing resources currently occupied by the device, h(t) represents the total computing resources owned by the device, and B(t) represents the normalized historical computing pressure.

[0065] Step 6: Construct a candidate set of D2D devices based on the Hurst index H of the D2D devices and the statistic S of the historical computing pressure of the D2D devices obtained by the Cox-Stuart test method;

[0066] Step 6-1: Calculate the Hurst index H of the D2D device based on the historical pressure information of the D2D device;

[0067] In this embodiment, the Hurst index H is also called the rescaled range method or R / S analysis, which is a measurement tool for characterizing the degree of fractal. The Hurst index reflects the autocorrelation of the time series, and especially reflects the long-term trend hidden in the series, which is statistically called long-term memory. When H = 0.5, it indicates that the computing pressure has a random walk and no fixed trend; when H < 0.5, it indicates that the computing pressure has a reverse trend to past changes and has anti-persistence, that is, the trend is more likely to reverse in the short term. At this time, the D2D devices with a decreasing computing pressure trend should be removed from the D2D device candidate set; when H > 0.5, it indicates that the computing pressure has a continuous trend to past changes and has persistence, that is, the trend is more likely to continue in the short term. At this time, the D2D devices with an increasing computing pressure trend should be removed from the D2D device candidate set.

[0068] In this embodiment, the Hurst exponent calculation method of the D2D device is as follows:

[0069] Assume that for a historical calculated pressure B(t), its length is T, and it is divided into n subsequences of length τ (where different subsequences have different lengths τ). For each subsequence, first obtain the increment of the historical calculated pressure ξ(t) and obtain τ increment values; then calculate the mean value E(ξ) of all historical calculated pressure increments. τ Next, calculate the range and standard deviation of all deviation history calculation pressures with t≤τ. Then, take the logarithm of the ratio of the range to the standard deviation. Finally, perform regression analysis on n τs to obtain the Hurst index H of the D2D device. The specific steps are shown in the following formula:

[0070] ξ(t)=B(t)-B(t-1)

[0071]

[0072] X(t,τ)=ξ(t)-E(ξ) τ

[0073] R(τ)=X(t,τ) max -X(t,τ) min

[0074]

[0075] Wherein, X(t,τ) represents the deviation sequence for all t≤τ, R(τ) represents the range of the deviation sequence, and S(τ) represents the standard deviation of the deviation sequence. It is the ratio relationship between the range and the standard deviation of the deviation sequence.

[0076] In this embodiment, for different H, as long as the following relationship is satisfied, the D2D device can be removed from the D2D device candidate set, thereby improving the service quality of the D2D device. When H<0.5, it indicates that in a future period of time, the computing pressure of the device has a large reverse trend. At this time, the D2D device with a decreasing trend in the past computing pressure should be excluded from the D2D device candidate set; when H>0.5, it indicates that in a future period of time, the computing pressure of the device has a high possibility of maintaining the original trend. At this time, the D2D device with an increasing trend in the past computing pressure should be removed from the D2D device candidate set.

[0077] Step 6-2: Use the Cox-Stuart test method to obtain the statistic S of the historical computing pressure of the D2D device, and judge the change trend of the historical computing pressure of the current D2D device according to the statistic S, which is specifically expressed as follows:

[0078] In this embodiment, take B(t) and B(t+c) to form a data pair. When the length of B(t) is an even number n, c can take If n is an odd number, then take Measure the increase or decrease by the positive or negative of B(t+c)-B(t). N + represents the number of increases, and N - represents the number of decreases. Calculate the historical computing pressure statistic of the D2D device When S is positive, the sequence may be increasing. When S is negative, the sequence may be decreasing, and the magnitude of S can represent the degree of increase or decrease; when 0.5<S<1, the time series is increasing and the trend is obvious, indicating that the computing pressure of the device has a strong increasing trend; when -1<S<-0.5, the time series is decreasing and the trend is obvious, indicating that the computing pressure of the device has a strong decreasing trend; when 0.2<S<0.5, the increasing or decreasing trend of the time series is weak; when S<0.2, the time series does not have an obvious trend.

[0079] Step 6-3: According to the Hurst index H and the statistic S obtained by the Cox-Stuart test method, exclude the D2D devices that satisfy: 0.5<S≤1, H>0.5 and -1<S≤-0.5, H<0.5 from the candidate set;

[0080] Step 7: Use the D2D device candidate set to assist the MEC network server for task offloading, and use the double-population genetic algorithm for iteration to obtain the optimal offloading strategy.

[0081] In this embodiment, the dual population genetic algorithm is iterated to obtain the optimal unloading strategy, such as Figure 2 As shown in the figure, first, the first generation of individuals is generated in the solution space by random generation; then, the fitness of individuals in the population is calculated according to the domain optimization objective, and the population is divided into two populations, the dominant population and the inferior population, according to the individual fitness. Different crossover rates and mutation rates are set for the dominant population and the inferior population, and independent crossover mutation is performed respectively; secondly, the fast non-dominated sorting algorithm is used to screen out the two populations entering the next iteration according to the scale; finally, the two populations are merged into the same population, and the above operations are repeated until the maximum number of iterations is reached. At this time, the individual with the highest fitness is selected and decoded to obtain the optimal unloading strategy.

[0082] In this embodiment, the crossover rate P of the dominant population is c and the mutation rate p m Specifically expressed as:

[0083]

[0084] Where, is the maximum value of the crossover rate, is the minimum value of the crossover rate, is the maximum value of the mutation rate, is the minimum value of the mutation rate, g is the current number of iterations, g max is the maximum number of iterations.

[0085] In this embodiment, the crossover rate of the inferior population is the maximum crossover rate of the superior population, which is expressed as The mutation rate of the inferior population is the maximum mutation rate of the superior population, which is expressed as

[0086] In this embodiment, the fast non-dominated sorting algorithm has two parameters for each individual i in the population: n(i) and s(i). n(i) represents the number of solutions that dominate solution i in the population, and s(i) represents the set of solutions dominated by solution i. The execution process is as follows:

[0087] (1) Find all individuals in the population with n(i) = 0 and store them in the set F(1).

[0088] (2) For each solution j in F(1), consider the set s(j) of solutions it dominates, and subtract 1 from n(k) for each individual k in s(j), that is, the number of solutions that dominate solution k is subtracted by 1.

[0089] (3) All individuals in F(1) are regarded as the first-level non-dominated individuals and are assigned the same non-dominated ranking order. The remaining individuals in the population are divided into non-dominated levels according to the above ranking method until all individuals are assigned a level.

[0090] (4) After performing a fast non-dominated sort, the F(2), F(2), ..., F(n-1) layers are placed into the next generation one by one until the population size exceeds the predetermined value after the F(n) layer is placed. At this time, individuals are selected from the F(n) layer according to their fitness and placed into the next generation until the population size is met.

[0091] 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 make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for offloading industrial Internet of Things resources based on Hurst and dual-population genetic algorithms, characterized by: The following steps are involved: Step 1: Build an industrial IoT resource offloading system model, including I D2D devices and N user-type devices; Step 2: Split the tasks generated by N user type devices into three parts and offload them to local computing, D2D device computing, and MEC server computing respectively; Step 3: Calculate the latency and energy consumption of the computation and offloading processes of the three parts after the split, and set the optimization goal P to minimize the total energy consumption and latency. Step 4: Obtain historical calculated pressure information of I D2D devices in the industrial Internet of Things; Step 5: Normalize all historical calculated pressure information of the D2D device; Step 6: Construct a candidate set of D2D devices based on the Hurst index H of the D2D devices and the statistic S of the historical computing pressure of the D2D devices obtained by the Cox-Stuart test method; Step 6-1: Calculate the Hurst index H of the D2D device based on the historical pressure information of the D2D device; Step 6-2: Use the Cox-Stuart test method to obtain the statistic S of the historical calculation pressure of the D2D device, and determine the change trend of the historical calculation pressure of the current D2D device based on the statistic S; Step 6-3: According to the Hurst index H and the statistic S obtained by the Cox-Stuart test method, as well as The D2D devices are excluded from the candidate set; Step 7: Use the D2D device candidate set to assist the MEC network server in task offloading, and use the dual-population genetic algorithm to iterate and obtain the optimal offloading strategy.

2. The industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithm according to claim 1 is characterized in that: The optimization goal P in step 3 is to minimize the total energy consumption and delay, which is expressed as minimizing the difference between the energy consumption and delay after overall offloading and the energy consumption and delay after only local offloading. The specific formula is as follows: ； Where, Indicates the total delay caused by the overall offloading of user type devices. Indicates the total energy consumption caused by overall unloading of user type devices. Indicates the delay incurred when the task is only offloaded locally. Indicates the energy consumption when the task is only offloaded locally. is the weight of the total delay, is the weight of total energy consumption.

3. The industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithm according to claim 2 is characterized in that: The Hurst index H described in step 6-1, when H = 0.5, indicates that the historical calculation pressure is a random walk with no fixed trend. In this case, the D2D device is retained in the D2D device candidate set; When H < 0.5, it indicates that the computing pressure has a reverse trend with respect to past changes and is anti-persistent, that is, the trend is more likely to reverse in the short term. In this case, D2D devices with a decreasing trend in historical computing pressure are removed from the D2D device candidate set. When H > 0.5, it indicates that the computing pressure has a continuous trend with respect to past changes and is persistent, that is, the trend is more likely to continue in the short term. In this case, D2D devices with an increasing trend in historical computing pressure are removed from the D2D device candidate set.

4. The industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithm according to claim 3 is characterized in that: The Cox-Stuart test method described in step 6-2 obtains the statistic S of the historical calculation pressure of the D2D device. When , it means that the historical computing pressure of the device has a strong increasing trend. , it indicates that the historical computing pressure of the device has a strong decreasing trend.

5. The industrial Internet of Things resource unloading method based on Hurst and dual-population genetic algorithm according to claim 4 is characterized in that: The dual-population genetic algorithm described in step 7 first generates the first generation of individuals in the solution space in a random manner; then, the fitness of the individuals in the population is calculated according to the domain optimization objective, and the population is divided into two populations, a dominant population and an inferior population, according to the individual fitness. Different crossover rates and mutation rates are set for the dominant population and the inferior population, and independent crossover mutation is performed on each population; secondly, a fast non-dominated sorting algorithm is used to select the two populations entering the next iteration according to their scale; finally, the two populations are merged into the same population, and the above operation is repeated until the maximum number of iterations is reached. At this time, the individual with the highest fitness is selected, and a decoding operation is performed on it to obtain the optimal unloading strategy.

6. The industrial Internet of Things resource offloading method based on Hurst and dual-population genetic algorithm according to claim 5 is characterized in that: Crossover rate of the dominant population and mutation rate Specifically expressed as: ； ； Where, is the maximum value of the crossover rate, is the minimum value of the crossover rate, is the maximum value of the mutation rate, is the minimum value of the mutation rate, is the current iteration number, is the maximum number of iterations; The crossover rate of the inferior population is the maximum value of the crossover rate of the superior population, which is expressed as , the mutation rate of the inferior population is the maximum mutation rate of the dominant population, expressed as .

Images