Method for partially unloading low-calculation-delay data of wireless power supply edge calculation system
The method for partial data offloading in wireless power edge computing systems addresses high computational delays by optimizing data offloading ratios and charging times, enhancing system performance and reducing latency.
Patent Information
- Application Number
- CN202411220892.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-12-06
AI Technical Summary
Traditional data offloading strategies in wireless power edge computing systems face high computational delays and resource pressures, especially in multi-user scenarios, due to the complexity of channel fading and frequent movement of edge devices, leading to performance degradation.
A method for partial data offloading with low computing latency, involving determining optimal data offloading ratios and charging times for wireless nodes, using a dynamic resource optimization algorithm to minimize system delay.
The method allows for flexible and efficient data unloading, achieving a global optimal solution for computing delay reduction by deriving the percentage of data to be unloaded from each node based on system state, outperforming binary offloading strategies.
Smart Images

Figure CN119095113A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of edge computing technology, and in particular to a method for partially unloading data with low computing latency in a wireless power supply edge computing system. Background Art
[0002] The ubiquitous Internet of Things (IoT) has become a key driver in today's wireless communications. However, IoT user nodes have inherent limitations, such as computing power, battery capacity, battery replacement, and overall life. The emergence of wireless power transfer (WPT) technology has successfully solved the problem of excessive battery energy consumption caused by operations such as data processing, alleviating the problems of insufficient battery energy supply and reduced battery life. In addition, multi-access edge computing (MEC) technology enables user node data to be transferred to dedicated edge servers for processing. This data transfer method is called data offloading. At present, there are two main ways of data offloading. One is binary offloading, that is, user nodes can only choose one of the two methods of "all data offloading" and "no data offloading"; the other is partial offloading, that is, user nodes can independently choose to offload part of the data to a dedicated edge server. MEC technology can effectively solve the problem of limited computing power of user nodes. Therefore, integrating WPT and MEC technologies into the wireless powered edge computing (WP-MEC) system can better solve the problems of limited energy and computing power of user nodes.
[0003] However, with the increasing number of user nodes, new challenges have been raised to the communication delay of the entire system. When solving the problem of minimizing the computational delay in multi-user scenarios, traditional non-online optimization algorithms are not suitable for making real-time offloading decisions when the channel fades rapidly due to their high algorithmic complexity. Data binary offloading strategies usually involve moving the entire task to the edge device for processing, resulting in high data transmission costs and high resource pressure on edge devices. If the edge device or mobile node moves frequently, the offloading and transmission of tasks may be disturbed, resulting in performance degradation or instability. Therefore, the current system's data processing speed and responsiveness can no longer meet the growing demand. Summary of the invention
[0004] Technical issues to be solved
[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method for partial data unloading with low computing delay for a wireless power supply edge computing system, which solves the technical problem of minimizing the computing delay of a wireless power supply edge computing system under a partial data unloading strategy for wireless nodes.
[0006] Technical Solution
[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:
[0008] In a first aspect, the present invention provides a method for partially unloading data with low computing latency in a wireless power supply edge computing system, comprising:
[0009] On the basis that the optimal value x of data offloading ratio of each wireless user node has been determined, the charging time t is set. 0 The initial value of A random number between 0 and 100;
[0010] Set the optimal charging time to And initially set up
[0011] At a given charging time t 0 Calculate the data unloading time of each node and local computation time Find the node with the longest local computing time among all local computing nodes and call it the worst local computing node WLC-WD, denoted by N w , and determine the value of the index value w of the WLC-WD node;
[0012] At a given charging time t 0 When Find all nodes The node with the largest value is called the system charging time lower bound node CTC-WD, denoted by N m , and determine the value of the index value m of the CTC-WD node, where p max N i The upper bound of the transmission power, N i represents the i-th WD, x i N i The proportion of data offloaded to AP, L i Represents N i The amount of data to be processed, W represents the transmission bandwidth of the uplink and downlink, g i Indicates that from N i The channel gain of the uplink to the AP, σ 2 represents the background noise variance; hi From AP to N i The channel gain of the downlink, P is the charging power, and μ∈(0,1) represents the energy collection efficiency of each WD;
[0013] calculate and
[0014] Compare and The size of determines the optimal charging time value of the system in the current time slot Based on The data partial uninstall is complete.
[0015] Optionally, calculate according to the following formula
[0016]
[0017] Calculate according to the following formula
[0018]
[0019] Among them, G i represents the number of clock cycles required for local computing to process one bit of data, f i represents the computation frequency of the local processor, p u N i transmission power.
[0020] Optionally, compare and The size of determines the optimal charging time value of the system in the current time slot include:
[0021] like but like but At this time if but like but like but
[0022] like but At this time, if but
[0023] like but like but At this time, if and but like and, but
[0024] in, K w (t 0 ) and K m (t 0 ), or ∈ i N i The computational energy efficiency factor of local computing, f max The upper limit of the local calculation frequency value,
[0026] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements a method for partially unloading data with low computing latency in a wireless power supply edge computing system as described in any one of the first aspects above.
[0027] In a third aspect, the present invention provides a storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, a method for partially unloading data with low computing latency in a wireless power supply edge computing system as described in any one of the first aspects above is implemented.
[0028] Beneficial Effects
[0029] The beneficial effects of the present invention are: a method for partial unloading of data with low computing delay in a wireless powered edge computing system of the present invention can derive the percentage of data that should be unloaded at each wireless node according to the actual state of the system. Compared with the data binary unloading strategy, this method can complete the data unloading process more flexibly and smoothly, and obtain the global optimal solution for the system computing delay. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 A schematic diagram of a system scenario provided by an embodiment of the present invention;
[0031] Figure 2 A graph showing the evolution of the training loss of a deep reinforcement learning network as a function of time slots provided by an embodiment of the present invention;
[0032] Figure 3 A schematic diagram of algorithm processing time provided by an embodiment of the present invention;
[0033] Figure 4A schematic diagram of system calculation delay in three different data offloading modes provided in an embodiment of the present invention;
[0034] Figure 5 A schematic diagram of system calculation delay in an offload priority model and a data full offload mode provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0035] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.
[0036] When solving the problem of minimizing computational delay in multi-user scenarios, traditional non-online optimization algorithms are not suitable for making real-time offloading decisions when the channel fades rapidly due to their high algorithm complexity.
[0037] Data binary offloading strategies usually involve moving the entire task to the edge device for processing, which results in high data transmission costs and heavy resource pressure on the edge device. Moreover, if the edge device or mobile node moves frequently, the offloading and transmission of the task may be disturbed, resulting in performance degradation or instability.
[0038] In view of the above shortcomings, the present invention designs a method for partial data unloading for the wireless powered edge computing system. The method can derive the percentage of data that should be unloaded from each wireless node according to the actual state of the system. Compared with the data binary unloading strategy, this method can complete the data unloading process more flexibly and smoothly, and obtain the global optimal solution for the system computing delay.
[0039] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.
[0040] In a first aspect, this embodiment provides a method for partially unloading data with low computing latency in a wireless power supply edge computing system, including:
[0041] On the basis that the optimal value x of data offloading ratio of each wireless user node has been determined, the charging time t is set. 0 The initial value of A random number between 0 and 100;
[0042] Set the optimal charging time to And initially set up
[0043] At a given charging time t 0 Calculate the data unloading time of each node and local computation time Find the node with the longest local computing time among all local computing nodes and call it the worst local computing node WLC-WD, denoted by N w , and determine the value of the index value w of the WLC-WD node;
[0044] At a given charging time t 0 When Find all nodes The node with the largest value is called the system charging time lower bound node CTC-WD, denoted by N m , and determine the value of the index value m of the CTC-WD node, where p max N i The upper bound of the transmission power, N i represents the i-th WD, x i N i The proportion of data offloaded to AP, L i Represents N i The amount of data to be processed, W represents the transmission bandwidth of the uplink and downlink, g i Indicates that from N i Channel gain of the uplink to the AP, σ 2 represents the background noise variance; h i From AP to N i The channel gain of the downlink, P is the charging power, and μ∈(0,1) represents the energy collection efficiency of each WD;
[0045] calculate and
[0046] Compare and The size of determines the optimal charging time value of the system in the current time slot Based on The data partial uninstall is complete.
[0047] Optionally, calculate according to the following formula
[0048]
[0049] Calculate according to the following formula
[0050]
[0051] Among them, G i represents the number of clock cycles required for local computing to process one bit of data, f i represents the computation frequency of the local processor, p u N i transmission power.
[0052] Optionally, compare and The size of determines the optimal charging time value of the system in the current time slot include:
[0053] like but like but At this time if but like but like but
[0054] like but At this time, if but
[0055] like but like but At this time, if and, but like and, but
[0056] in, K w (t 0 ) and K m (t 0 ), or ∈ i N i The computational energy efficiency factor of local computing, f max The upper limit of the local calculation frequency value,
[0058] The following is a specific embodiment to further illustrate a method for partially unloading data with low computing latency in a wireless power supply edge computing system of the present application:
[0059] Reference Figure 1 In the wireless power supply edge computing system scenario to which the present invention is applicable, there is one access node (AP, Access Point), which communicates data with an edge computing server through a wired manner, and N wireless user nodes (WD, Wireless Node). i represents the i-th WD, and g i and h i Respectively represent from N i The channel gain of the uplink and downlink to the AP. In each time slot T, the AP can transmit energy to all WDs at the same time, and the energy transmission time is t 0 Indicates that the transmission power is P, and the energy collection efficiency of each WD is represented by μ∈(0,1). Then N i The received energy can be expressed as After the AP completes powering the WDs, all WDs can use the received energy to process their own data. i Represents N i The amount of data to be processed, x i N i The ratio of data offloaded to AP is N i The amount of data unloaded is x i L i , the amount of data calculated locally is (1-x i )L i , then N i Data unloading time and local computation processing time It can be expressed as follows:
[0060]
[0061]
[0062] Among them G i represents the number of clock cycles required for local computing to process one bit of data, W represents the transmission bandwidth of the uplink and downlink, σ 2 represents the background noise variance, p i N i The transmission power, p max N i The upper bound of the transmission power, f i represents the computation frequency of the local processor, f max represents f i The upper bound of i N i The computational energy efficiency coefficient of local computing.
[0063] The communication scenario to which the present invention is applicable is that the local computing power of the wireless user node is much smaller than the computing power of the edge computing server, so the computing delay of the edge computing server can be ignored. And the data communication delay between the access node and the edge computing server connected to it is negligible. In addition, the data unloading and local computing of each wireless user node can be performed in parallel, but the data unloading between each wireless user node is performed in time division multiple access (TDMA), that is, only one wireless user node can perform data unloading at the same time.
[0064] The system targeted by the present invention has the following features: (1) Zero tolerance for data discard. That is, in any time slot T, the data of each WD must be processed completely, and no unprocessed data is allowed. (2) Maximization of data offloading power. Since the computing delay of the computing server is negligible, the wireless user node will give priority to using energy to offload data to the edge computing server. That is, N i , The node needs to use the maximum power p i =p max After completing the partial data offloading, the remaining energy will be used to complete the local calculation of the remaining data.
[0065] Then N i The energy consumed by data offloading and local computing can be expressed as:
[0066]
[0067]
[0068] In summary, the computational delay T of the entire system in each time slot is delay It can be expressed as:
[0069]
[0070] Then the computational delay minimization problem P0 of the entire system can be expressed as:
[0071]
[0072]
[0073]
[0074]
[0075] 0≤t 0 (10)
[0076] where x = {x 1 ,...,x N}, f={f 1 ,...,f N}.
[0077] To minimize the computational delay of the entire wireless power edge computing system in the current time slot, the first step is to determine the optimal value x of the data offloading ratio of each wireless user node; then the second step is to further obtain the optimal charging time t based on the determined data offloading ratio. 0 , optimal local computation frequency F. According to the above two steps, P0 is decomposed into a main problem P1 and a sub-problem P2, as shown below.
[0078] (P1):min x T delay (11)
[0079]
[0080]
[0081]
[0082]
[0083] 0≤t 0 (16)
[0084] The main problem P1 is solved by using the data offloading dynamic resource optimization algorithm to obtain the optimal value x of the data offloading ratio of each wireless user node. Based on the determination of the optimal x, the solution process of sub-problem P2 is as follows.
[0085] For each node, And when F i ∈(0,+∞) According to formula (4), we can find the corresponding local calculation frequency value so that That is, the remaining energy after data unloading is completely consumed, so that the local computing frequency value is maximized, thereby minimizing the local computing time of the node. It can be inferred from this According to formula (2), we can infer The sum of the charging time and local computing time of each node in time slot T is expressed as function Q i (t 0 ),but
[0086]
[0087] For the convenience of expression, let:
[0088]
[0089]
[0090]
[0091] Remember Q i (t 0 ) corresponds to the minimum value of t 0 Value is Q i (t 0 ) Taking the derivative and setting it to 0 gives
[0092]
[0093] However, in actual situations, there must be an upper limit to the local calculation frequency value, denoted as f max Based on the above, when When, according to formula (4), the maximum energy consumed locally is Then when When That is N i The energy obtained in time slot T cannot be completely consumed in the current time slot; when hour, For the convenience of description,
[0094]
[0095] Therefore, in practice, the function Q i (t 0 ) at t 0 ∈[t 0 ′ ,+∞) will become a linear function. 0 How to increase it? The energy consumed by local computing has reached its maximum value, and f i =f max , then according to formula (2), the minimum value that the local computing time can reach will not change. i (t 0 ) represents the function Q in actual situation i (t 0 ),but
[0096]
[0097] In order to obtain K i (t 0 ) needs to be discussed in categories: When t 0 =t 0 ′ Can make Ki (t 0 ) obtains the minimum value; when season Can make K i (t 0 ) to obtain the minimum value. For the convenience of subsequent description, let K i (t 0 )T when the minimum value is obtained 0 value:
[0098]
[0099] At a given charging time t 0 When the local computing time of all local computing nodes is the longest, the node is called the worst local computing wireless device (WLC-WD), denoted by N w , w is the index value of the WLC-WD node.
[0100] At a given charging time t 0 When The node with the largest value is called the charging time constrained wireless device (CTC-WD), denoted by N m , m is the index value of the CTC-WD node.
[0101] Since all nodes' local computations are performed in parallel, the system latency will be directly affected by The maximum value of the influence. For a given t 0 ,make The local computing time value of the node with the longest local computing time can be adjusted by adjusting t 0 The value of can be obtained The minimum value of express The minimum value of t 0 value. There are several different situations for functions, which are classified and analyzed below:
[0102] CASE1:
[0103] In this case The function value is exactly equal to If a function in takes a value, then according to formula (24) we can know
[0104] CASE2:
[0105] In this case, further classification analysis is required:
[0106] CASE2-1 at this time The feasible domain is Right now Unable to obtain, and due to mean By N w and N m K corresponding to the two nodes w (t 0 ), K m (t 0 ) together. At this time, the minimum point should be K w (t 0 ) and K m (t 0 ), the intersection point corresponds to t 0 use Said that according to And formula (23) can be inferred that K w (t 0 ), K m (t 0 ) in functional form, and let K w (t 0 )=K m (t 0 ) can be obtained:
[0107]
[0108] at this time
[0109] CASE2-2: This situation can be further divided into two situations:
[0110] CASE2-2-1:
[0111] At this time, the minimum point is K w (t 0 ) and K m (t 0 ), according to And formula (23) can be inferred that K w (t 0 ), K m (t 0 ) in functional form, and let K w (t 0 )=K m (t 0 ) can be obtained at this time Satisfies formula (25), that is
[0112] CASE2-2-2:
[0113] At this time, the minimum point is K w (t 0 ) and K m (t 0 ), according to And formula (23) can be inferred that K w (t 0 ), K m (t 0 ) in functional form, and let K w (t 0 )=K m (t 0 ) can be obtained
[0114]
[0115] CASE3:
[0116] In this case The minimum value is equal to K w (t 0 ), exist In the feasible region of
[0117] After the above analysis, we have initially obtained The value of and It has also been tentatively determined. Next, based on the comparison and The value of can determine whether to continue looking for a better Values:
[0118] ①
[0119] In this case, The minimum value has been obtained for t 0 Any adjustment will increase the system calculation delay, so Already t 0 The best value that can be obtained.
[0120] ②
[0121] In this case, Determines the overall computing delay of the system, so if we continue to reduce The value of can further reduce the overall computing delay of the system.
[0122] The sum of the charging time in time slot T and the data unloading time of all nodes is recorded as h(t 0 ):
[0123]
[0124] Based on what is described in ②, the above-mentioned CASE1, CASE2, and CASE3 can further explore the optimal value.
[0125] CASE1':
[0126] Based on the optimization of CASE1, When h(t 0 )exist The above are monotonically increasing. exist are monotonically decreasing and monotonically increasing respectively. So when t 0 When h(t 0 ) decreases, Increase, assuming hour At this time, if we continue to reduce t 0 Will cause It can be inferred that t 0 It is definitely not the optimal value.
[0127] Therefore, Available At this time, t 0 The optimal value of
[0128]
[0129] CASE2':
[0130] Based on the optimization of CASE2, When
[0131] When WLC-WD changes, the original N m Became the new N w , according to the derivation process in CASE1', the optimal Should be:
[0132]
[0133] CASE3':
[0134] Based on the optimization of CASE3, When Shi Ruo but otherwise
[0135] Figure 2 Figure 2. The evolution of the training loss of a deep reinforcement learning network over time slots. Initially, the training loss experiences a rapid decline from 0 to 5000. This decline indicates that the algorithm is learning and improving rapidly. However, after reaching the 5000th time slot, the training loss begins to converge and oscillate. These oscillations are likely due to random sampling of the training data. This convergence phase indicates that the algorithm has reached a point of diminishing returns.
[0136] Figure 3 The processing time of the algorithm is shown. The average latency of the algorithm is 4.01e-04, which provides a key reference point for evaluating system performance.
[0137] Figure 4 and Figure 5 The overall computing delay of the system under three different data offloading modes was compared: full data offloading mode (all data is offloaded to the edge computing server), full local computing mode (all data is kept locally for calculation) and offloading priority mode (partial data offloading mode adopted by the present invention).
[0138] Figure 4 In the , all-local computing mode shows significantly higher latency compared to the other two modes. This difference highlights the advantage of offloading computing tasks in reducing latency.
[0139] Figure 5 In the data full offloading mode, the overall computing delay is higher than that of the offloading priority mode, and the overall computing delay of the system in the offloading priority mode changes dynamically with the time slot and finally converges to an optimal delay value.
[0140] In a second aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements a method for partially unloading data with low computing latency in a wireless power supply edge computing system as described in any one of the first aspects above.
[0141] In a third aspect, an embodiment of the present invention provides a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, a method for partially unloading data with low computing latency in a wireless power supply edge computing system as described in any one of the first aspects above is implemented.
[0142] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0143] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.
[0144] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.
Claims
1. A method for partially unloading data with low computing latency in a wireless power supply edge computing system, characterized in that: include: On the basis that the optimal value x of the data offloading ratio of each wireless user node has been determined, the initial value of the charging time t0 is set, which is recorded as A random number between 0 and 100; Set the optimal charging time to And initially set up Given the charging time t0, calculate the data unloading time of each node and local computation time Find the node with the longest local computing time among all local computing nodes and call it the worst local computing node WLC-WD, denoted by n w , and determine the value of the index value w of the WLC-WD node; Given the charging time t0, calculate Find all nodes The node with the largest value is called the system charging time lower bound node CTC-WD, denoted by N m , and determine the value of the index value m of the CTC-WD node, where p max N i The upper bound of the transmission power, N i represents the i-th WD, x i N i The proportion of data offloaded to AP, L i Represents N i The amount of data to be processed, W represents the transmission bandwidth of the uplink and downlink, g i Indicates that from N i The channel gain of the uplink to the AP, σ 2 represents the background noise variance; h i From AP to N i The channel gain of the downlink, P is the charging power, and μ∈(0,1) represents the energy collection efficiency of each WD; calculate and Compare and The size of determines the optimal charging time value of the system in the current time slot Based on The data partial uninstall is complete.
2. According to claim 1, a method for partially unloading data with low computing latency in a wireless power supply edge computing system is characterized in that: Calculate according to the following formula Calculate according to the following formula Among them, G i represents the number of clock cycles required for local computing to process one bit of data, f i represents the computation frequency of the local processor, p i N i transmission power.
3. A method for partially unloading data with low computing latency in a wireless power supply edge computing system according to claim 2, characterized in that: Compare and The size of determines the optimal charging time value of the system in the current time slot include: like but like but At this time if but like but like but like but At this time, if but like but like but At this time, if and, but like and, but in, K w (t0) and K m The intersection point of (t0), or ∈ i N i The computational energy efficiency factor of local computing, f max The upper limit of the local calculation frequency value, 4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a method for partial unloading of data with low computing latency in a wireless power supply edge computing system as described in any one of claims 1 to 3.
5. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that: When the processor executes the computer program, it implements the low computing latency data partial unloading method of a wireless power supply edge computing system as described in any one of claims 1 to 3.
