Cloud edge computing power regulation and control method and system based on edge computing

Abstract

The invention relates to the technical field of power systems, in particular to a cloud edge computing power regulation and control method and system based on edge computing, and the method comprises the following steps: S1, collecting the resource state data of a cloud data center, edge nodes and terminal equipment in real time; s2, constructing a dynamic scheduling strategy based on a deep reinforcement learning model, and generating a computing power allocation scheme according to task types, priorities, real-time resource states and historical load data; s3, performing hierarchical processing on the task; s4, autonomously executing emergency task scheduling in a network disconnection or high-delay scene through a local autonomous module of the edge node; s5, global energy consumption is optimized based on a genetic algorithm, the steps S1 to S5 cooperatively achieve dynamic regulation and control of cloud edge computing power, and accurate input is provided for dynamic scheduling by collecting multi-dimensional resource data in real time; through a task layering mechanism, it is ensured that high-real-time tasks are preferentially processed locally, and end-to-end delay is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a cloud-edge computing power control method and system based on edge computing. Background Art

[0002] With the rapid development of the Internet of Things (IoT), 5G communications, and artificial intelligence technologies, the data generated by massive terminal devices is growing exponentially. The traditional cloud computing model, which transmits all data to the cloud for processing, has gradually exposed problems such as high latency, large bandwidth usage, and privacy leaks. Edge computing effectively reduces data transmission latency and improves real-time performance by sinking computing, storage, and network resources to edge nodes close to terminal devices, becoming a key technical approach to addressing these issues. However, in practical applications, the coordinated scheduling of edge nodes and cloud resources still faces many challenges.

[0003] In the existing technology, the computing power allocation of cloud-edge collaboration mostly adopts static or semi-dynamic strategies. For example, some solutions allocate tasks to nodes of a fixed level based on preset rules, or adjust resource allocation according to historical load data. This type of method is difficult to adapt to dynamically changing network environments and task requirements, especially in scenarios with bursty traffic, coexistence of heterogeneous devices, and large differences in task priorities, which can easily lead to edge node overload or idle cloud resources. In addition, the performance of existing algorithms in terms of resource utilization optimization, energy consumption control, and task scheduling efficiency is still insufficient. For example, traditional load balancing algorithms (such as polling and weighted minimum connections) do not fully consider the real-time computing power status of edge nodes and the dependency between tasks, which may cause local hotspots; and although the dynamic scheduling model based on deep learning has a certain degree of adaptability, it has problems such as high model training cost and insufficient real-time response capability.

[0004] On the other hand, existing systems have shortcomings in the intelligence and flexibility of computing power regulation. Most systems use a centralized decision-making mechanism and rely on unified scheduling in the cloud, which limits the autonomy of edge nodes and makes it difficult to respond quickly in disconnected or high-latency environments. At the same time, the coupling between resource allocation strategies and business needs is low, and there is a lack of differentiated regulation capabilities for different application scenarios (such as industrial Internet of Things, autonomous driving, and smart grids). For example, in smart grids, real-time monitoring and control tasks have extremely high requirements for low latency and high reliability, while data analysis tasks can tolerate a certain degree of delay, but existing systems are often unable to efficiently distinguish and optimize the scheduling of such heterogeneous tasks.

[0005] Furthermore, the heterogeneity and varying computing power of edge nodes further complicate regulation. Edge devices from different vendors exhibit significant differences in computing power, storage capacity, and communication protocols. Existing technologies lack a unified resource abstraction and dynamic adaptation mechanism, resulting in inefficient cross-platform resource scheduling. Regarding energy management, existing solutions focus primarily on optimizing energy consumption at a single node, while neglecting the global energy optimization required by cloud-edge collaboration, making it difficult to meet the demands of green computing.

[0006] Therefore, there is an urgent need for a new control method and system that can achieve dynamic coordination of cloud-edge computing power, adaptive resource allocation, low latency and high reliability, so as to solve the problems of low resource utilization, high response delay, and insufficient energy consumption optimization in existing technologies and meet the diverse needs in complex scenarios. Summary of the Invention

[0007] The purpose of the present invention is to provide a cloud-edge computing power control method and system based on edge computing to solve the problems existing in the above-mentioned prior art.

[0008] The above technical objectives of the present invention are achieved through the following technical solutions:

[0009] A cloud-edge computing power control method based on edge computing includes the following steps:

[0010] Step S1: Real-time collection of resource status data of cloud data centers, edge nodes, and terminal devices, including computing power utilization, storage capacity, network bandwidth, task queue length, and energy consumption indicators;

[0011] Step S2: Build a dynamic scheduling strategy based on the deep reinforcement learning model to generate a computing power allocation plan based on task type, priority, real-time resource status, and historical load data;

[0012] Step S3: Perform hierarchical processing on tasks, prioritize high-real-time tasks to edge nodes, and dynamically migrate computationally intensive tasks to the cloud or adjacent edge node clusters;

[0013] Step S4: Through the local autonomous module of the edge node, emergency task scheduling is autonomously performed in the scenario of network disconnection or high latency, and the global resource status of the cloud is synchronously updated;

[0014] Step S5: Optimize global energy consumption based on genetic algorithms, combine the energy consumption data of cloud and edge nodes, and dynamically adjust the start and stop status of computing nodes and task allocation weights.

[0015] By adopting the above technical solution, steps S1 to S5 collaboratively realize the dynamic regulation of cloud-edge computing power: step S1 provides accurate input for dynamic scheduling by collecting multi-dimensional resource data in real time; step S2 dynamically generates allocation plans based on the TD3 algorithm to adapt to network fluctuations and changes in task priorities; step S3 ensures that high-real-time tasks are processed locally first through the task layering mechanism, reducing end-to-end delay; the autonomous module of step S4 maintains the edge node function in the network disconnection scenario to ensure the execution of critical tasks; step S5 combines genetic algorithms to optimize global energy consumption and balance computing load and energy costs.

[0016] In a further embodiment, the deep reinforcement learning model in step S2 adopts a dual-delay deep deterministic policy gradient TD3 algorithm, whose state space includes node computing power, task delay constraints and network topology relationships, the action space is the task allocation decision vector, and the reward function is composed of task completion time, resource utilization and energy consumption weights.

[0017] By adopting the above technical solutions, the TD3 algorithm reduces the problem of strategy over-estimation through a dual-critic network, combines node computing power and network topology status to generate optimal task allocation decisions, reduces scheduling delays by 15%-20%, and increases resource utilization to more than 85%.

[0018] In a further embodiment, step S3 further includes: dividing the real-time level according to the service level agreement of the task, adopting a preemptive scheduling strategy for industrial control tasks, and adopting a flexible resource reservation mechanism for data analysis tasks.

[0019] By adopting the above technical solutions, the SLA-based task classification mechanism marks industrial control tasks as the highest priority, and ensures millisecond-level response through preemptive scheduling; data analysis tasks use elastic resource pools to dynamically allocate computing power to avoid resource waste.

[0020] In a further embodiment, the local autonomous module in step S4 has a built-in lightweight decision-making model, which regularly synchronizes parameters with the cloud model through federated learning, supporting edge nodes to perform short-term prediction and task scheduling based on local data in an offline state.

[0021] By adopting the above technical solutions, the SLA-based task classification mechanism marks industrial control tasks as the highest priority, and ensures millisecond-level response through preemptive scheduling; data analysis tasks use elastic resource pools to dynamically allocate computing power to avoid resource waste.

[0022] The present invention also discloses a cloud-edge computing power control system based on edge computing, comprising:

[0023] Resource monitoring unit, used to collect resource status data of the cloud, edge nodes and terminal devices in real time;

[0024] Dynamic scheduling engine, integrating deep reinforcement learning models to generate and optimize computing power allocation plans;

[0025] Task hierarchical processor, which prioritizes tasks according to their type and SLA and assigns them to corresponding nodes;

[0026] The edge autonomy module is deployed on each edge node to support emergency task scheduling in an offline environment;

[0027] Energy consumption optimization unit, which implements node start and stop strategy with global energy consumption optimization based on genetic algorithm.

[0028] By adopting the above technical solution, the main scheduler coordinates global resource allocation, and the sub-scheduler responds quickly based on local load, shortening the task scheduling time to less than 50ms.

[0029] In a further embodiment, the dynamic scheduling engine adopts a distributed architecture, including a cloud-based main scheduler and edge sub-schedulers. The main scheduler is responsible for global resource coordination, and the sub-scheduler quickly responds to task requests based on local status.

[0030] In a further embodiment, the task hierarchical processor supports multi-protocol adaptation, including MQTT, CoAP and custom industrial protocols, and abstracts the computing resources of heterogeneous edge devices through virtualization technology.

[0031] By adopting the above technical solutions, the protocol adaptation interface supports industrial protocols such as MQTT, CoAP and Modbus. Through virtualization technology, the computing power of heterogeneous devices is uniformly abstracted into a schedulable resource pool, and the cross-platform scheduling efficiency is improved by 30%.

[0032] In a further embodiment, the energy consumption optimization unit introduces a carbon emission factor, combines the real-time electricity price of the regional power grid and the proportion of clean energy, and dynamically adjusts the load distribution of the computing nodes. The carbon emission factor needs to be calculated according to the international standard ISO14064 to dynamically adjust the node load distribution.

[0033] By adopting the above technical solutions, the protocol adapter interface supports industrial protocols such as MQTT, CoAP and Modbus. Through virtualization technology, the computing power of heterogeneous devices is uniformly abstracted into a schedulable resource pool, and the cross-platform scheduling efficiency is improved by 30%. Virtualization technology includes Docker containers and Kubernetes cluster management to realize resource pooling of heterogeneous devices.

[0034] In summary, the present invention has the following beneficial effects:

[0035] 1. Dynamic control of cloud-edge computing power is achieved through the collaborative efforts of steps S1 to S5: Step S1 collects multi-dimensional resource data in real time to provide precise input for dynamic scheduling; Step S2 dynamically generates allocation plans based on the TD3 algorithm to adapt to network fluctuations and changes in task priorities; Step S3 ensures that high-real-time tasks are processed locally first through a task layering mechanism, reducing end-to-end latency; Step S4's autonomous module maintains edge node functionality in network outages to ensure the execution of critical tasks; Step S5 combines genetic algorithms to optimize global energy consumption, balancing computing load and energy costs.

[0036] 2. Industrial control tasks are marked as the highest priority through an SLA-based task classification mechanism, and millisecond-level response is ensured through preemptive scheduling. Data analysis tasks use an elastic resource pool to dynamically allocate computing power to avoid resource waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a system architecture diagram of a cloud-edge computing power control system based on edge computing of the present invention;

[0038] Figure 2 This is an architecture diagram of the household information collection system of the cloud-edge computing power control system based on edge computing of the present invention;

[0039] Figure 3 It is a flow chart of a cloud-edge computing power control method based on edge computing of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described in detail below with reference to the accompanying drawings.

[0041] The same parts are denoted by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the attached Figure 1 In the description, the terms "bottom" and "top," "inner" and "outer" refer to directions toward or away from a particular component geometry, respectively. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, features defined as "first" or "second" may explicitly or implicitly include one or more of such features. Throughout this specification, "plurality" means two or more, unless otherwise specifically defined in terms of the center's direction.

[0042] Example 1:

[0043] like Figure 1-Figure 3As shown, a cloud-edge computing power control method based on edge computing is described below. The direction of power dispatching is explained below. With the increasing application of new energy technologies and distributed energy access technologies, the development of power Internet of Things services such as real-time monitoring of power system equipment, electric vehicle charging and discharging, and automatic bill payment has increasingly increased the requirements for PLC communication speed. Currently, advanced power line carrier products mainly use OFDM modulation technology. The anti-interference ability and speed of PLC communication have been greatly improved, and the scope of application has expanded accordingly. However, it is still difficult to meet telecommunications-grade standards in terms of communication quality and speed. Based on this, the present invention combines quantum coding theory and genetic algorithm principles based on the service access requirements within the substation area and the channel difference characteristics of power line carriers to propose a broadband power line OFDM subcarrier allocation principle based on quantum genetic algorithm. OFDM divides the channel into multiple subcarrier channels and adopts a dynamic resource allocation method. The allocated frequency and subcarrier modulation method are selected according to the channel quality. The channel quality is graded. High modulation is used for good quality and low modulation is used for poor quality. Channels with extremely poor quality are not used. With the advent of quantum computing theory, modulation systems will benefit from the development of quantum genetic algorithms (QGAs). These algorithms use quantum bits to represent information. Due to the randomness of measurement, QGAs expand the search range of populations and rapidly approach the optimal solution based on the direction of genetic evolution. QGAs require fewer populations and iterations than traditional genetic algorithms, minimizing the probability of falling into local optimal solutions. Therefore, QGAs are suitable for optimizing subcarrier allocation in broadband power line OFDM systems.

[0044] This includes subcarrier allocation, and the subcarrier allocation method is as follows:

[0045] The quantum genetic algorithm can be used to optimize the subcarrier allocation of the model. The algorithm process is as follows:

[0046] Step 1: Initialize the population and encode the individuals. Suppose a system has N subcarriers and K users, each of which contains N chromosomes, corresponding to N subcarriers to be allocated. A chromosome contains S = [log2K] genes, representing the user assigned to this subcarrier. The encoded individual is represented as:

[0047]

[0048] in: It represents the jth individual of the tth generation, and from left to right, each interval is represented as the first subcarrier, the second subcarrier...the Nth subcarrier.

[0049] Step 2: Quantum state measurement, generate a random variable between 0 and 1, and use this random variable to measure the quantum superposition state, so that the quantum superposition state collapses from one state to another specific state.

[0050] Step 3: Individual evaluation: Each state obtained by measurement is individually evaluated. The quality of individual states can be quantitatively expressed using an appropriate objective function to accelerate convergence. The resource allocation problem based on proportional fairness is a multi-objective optimization problem. In multi-objective optimization problems, there is generally no solution that optimizes multiple objectives simultaneously, and feasible solutions are often used. A feasible solution is a compromise that tries its best to meet the requirements of multiple objectives. Therefore, an evaluation function that comprehensively considers fairness and rate is proposed:

[0051]

[0052] In the above formula, the left side represents the total sending rate, and the right side represents the system service fairness index (SFI). The SFI ranges from 0 to 1, and proportional fairness increases from 0 to 1. Since a smaller evaluation function generally indicates greater suitability, the negative sign is added.

[0053] Step 4: Evolution. When the algorithm makes the subcarriers to be allocated reach the maximum number of iterations or meets the system requirements, the algorithm terminates; otherwise, the offspring continues to evolve until f(b i ), that is, the optimal individual of this generation; non-optimal individuals evolve towards the optimal individual according to this evolutionary rule.

[0054] Step 5: The evolved individuals return to Step 2 and become a new population to continue evolving.

[0055] It also includes algorithm optimization, the optimization methods are as follows:

[0056] Using QGA reduces the population size, selects the evolutionary direction, and accelerates convergence. However, PLC communications have a large number of subcarriers. The HomePlug AV standard has 917 available subcarriers, with communication frequencies ranging from 1 MHz to 34 MHz. As the number of subcarriers increases, the number of qubits representing each individual qubit also increases. Excessive variables to optimize significantly increases algorithm complexity. Furthermore, powerline channels exhibit certain regularities, with local line topology and access point locations being largely fixed. Therefore, identifying and leveraging these regularities can optimize the algorithm and improve its efficiency.

[0057] Carrier clustering: Multiple subcarriers in a channel are grouped together as a whole to form a subband. During allocation, subcarriers in the same subband are adjusted using the same modulation scheme and assigned to the same user, effectively reducing the number of units to be allocated and the signaling overhead incurred when using feedback modulation. Subbands can be divided into equal or uneven bandwidths. In equal bandwidth division, each subband has the same number of subcarriers, while in uneven bandwidth division, the number of subcarriers in each subband is determined and allocated based on the real-time channel status.

[0058] Multiple subcarriers are aggregated into one subband. To ensure the bit error rate of information transmission, the subband channel is determined by the subcarrier with the worst quality. When the subband is too large, the system performance will degrade. When L subcarriers are aggregated into one subband, the number of resource allocations is reduced to N / L. For quantum genetic algorithms, the most complex process is the sorting of target evaluation. The complexity of using quick sort is O(Nlog2N). After subband division, the complexity is reduced to In addition, sub-band division also compresses the search space of the algorithm. Before sub-band division, the search space is K N , after dividing the subbands, the search space is reduced to The reduction of the search space also reduces the number of iterations and population size required by the algorithm. Therefore, sub-band division greatly reduces the complexity of the algorithm.

[0059] Population optimization: Individual selection plays a crucial role in the rapid convergence of genetic algorithms. Therefore, adding high-performing individuals to the population can significantly improve algorithm efficiency. In power line networks, load and network topology changes over time exhibit regular patterns. Power line channel noise can be categorized as short-term periodic noise and long-term sudden noise. Short-term periodic noise is primarily caused by electrical appliances on the line. Whether periodic impulse noise triggers resource reallocation depends on the strength of the noise. Sudden noise is primarily caused by plugging and unplugging appliances or poor connections. This type of noise has a significant impact and generally requires resource reallocation. Therefore, it is important to record resource allocation results after power line channel mutations over a period of time. A self-learning module can also be used to learn and analyze resource allocation statistics to understand the patterns of channel variation. Recording the allocation results after channel mutations identifies high-performing individuals and adds them to the population to improve population quality.

[0060] OFDM subcarrier frequency aggregation based on a combinatorial optimization algorithm: With the rapid development of power line carrier communication technology, the spectrum bandwidth required for new communication services is increasing. The current shortage of spectrum resources is becoming increasingly prominent, gradually becoming one of the most scarce resources in today's society. Under such a spectrum situation, in order to fully utilize spectrum resources, improve spectrum efficiency, provide higher transmission rates, and alleviate the current spectrum shortage, effective measures must be taken to overcome the bottlenecks caused by spectrum resource shortages and harsh transmission environments in the development of power line communication. In addition, in the new energy microgrid environment, power lines are subject to severe noise interference in certain frequency bands, which seriously affects signal transmission. To avoid these frequency bands and achieve higher transmission data rates, frequency aggregation technology can improve the communication performance of new energy microgrids.

[0061] Shannon's theorem states that, given a given system bandwidth, the achievable transmission rate has an upper limit. Therefore, while improving existing technologies and introducing new ones, the inevitable trend in the development of power line communication technology is to adopt flexible and effective methods to fully utilize limited spectrum resources. Spectrum aggregation technology is an effective means of achieving broadband transmission when spectrum resources are insufficient.

[0062] Traditional spectrum management methods can no longer meet the current demand for spectrum resources in the information age. Future spectrum resource management and allocation will be more flexible and market-oriented. As the market value of spectrum continues to increase, the proper use of spectrum will be determined by market competition, and the era when spectrum will ultimately become a commodity is rapidly approaching. Spectrum aggregation technology can "bundle" multiple idle frequency bands or spectrum fragments for simultaneous use, providing broadband services for communication systems in situations where large, continuous bandwidth is unavailable. The system allocates five consecutive subcarriers to a single user to achieve higher peak rates.

[0063] Spectrum aggregation can be divided into two categories: continuous spectrum aggregation and non-continuous spectrum aggregation. Non-continuous spectrum aggregation can be divided into intra-band and inter-band. The present invention can implement a spectrum aggregation method based on non-continuous OFDM at the physical layer. The bandwidth of the entire frequency band is B, which is divided into N subcarriers. The bandwidth of each subcarrier is B / N. The total frequency band occupied by these N subcarriers is continuous. However, in power line communication, some frequency bands have serious noise interference, which makes it impossible to transmit signals on these frequency bands. In order to realize OFDM modulation under discrete spectrum conditions, the non-continuous OFDM derived from the OFDM system is used for reference. The multi-carrier characteristics of OFDM divide the broadband carrier into several narrowband subcarriers, and then by shutting down some of the subcarriers, a series of non-continuous subcarriers are formed. These non-continuous subcarriers can be used to realize data transmission. OFDM is particularly suitable for the physical layer transmission scheme of discrete frequency band spectrum aggregation.

[0064] In spectrum aggregation, understanding how users utilize idle resources and optimizing one objective while ensuring another is a typical combinatorial optimization problem. Given the characteristics of spectrum aggregation, the knapsack and bin packing problems in combinatorial optimization are best suited to solving this objective optimization problem.

[0065] In power line systems, when authorized users are absent, cognitive users can utilize spectrum aggregation technology to occupy multiple idle frequency bands for transmission, thereby increasing data rates. These idle frequency bands can be continuous or discrete. Therefore, from the perspective of cognitive users, efficiently utilizing these idle frequency bands is a matter of optimal combination. The optimal combination involves mathematically combining discrete events according to the optimization objective to achieve the desired effect. Based on spectrum aggregation technology, the optimal combination of discrete idle frequency bands is selected to serve cognitive users, thereby improving spectrum utilization.

[0066] The two optimization objectives of the spectrum aggregation algorithm are achieved through optimal combination theory. The so-called optimal combination is to rationally combine discrete events to obtain the optimal solution to the objectives, while satisfying given conditions. The mathematical model of the combinatorial optimization problem can be expressed as follows:

[0067] minf(a)or maxf(a)

[0068] stg(a)≥0

[0069] a∈D

[0070] Here, x is the decision variable, the objective function f(a) can be either a maximum or a minimum, and the decision variable is subject to the constraints of the function g(a). D represents the set of decision variables. When (D, F, f) represents the optimal combination strategy, the domain can be represented by D. If F = {a|a∈D, g(a)≥0} describes all feasible solutions, then every element in the set F is a solution to the optimal combination problem. A feasible solution a* where the objective function f satisfies f(a*) = min{f(a), a∈F} or f(a*) = max{f(a), a∈F} is called the optimal solution to the problem.

[0071] Typical combinatorial optimization problems include the knapsack problem, the traveling salesman problem, the integer linear programming problem, the bin packing problem, and the constrained machine sorting problem. The knapsack problem is used here.

[0072] The classic description of the knapsack problem is as follows: Suppose there are j objects, where j is an integer between 1 and n. Each object has its corresponding value vj and its corresponding volume bj. The capacity of the entire knapsack is represented by Y. Then the knapsack problem is how to load objects with the maximum total value without exceeding the capacity of the knapsack. Its strict mathematical description is as follows:

[0073]

[0074] a j ∈ {0, 1} j = 1, 2, …, n

[0075] Let aj represent whether the j-th object is selected. If aj = 1, it means the object is selected and loaded into the knapsack. If aj = 0

[0076] it means the object is not selected to be loaded. The function f(a1, a2, …, an) is used to calculate the total value of the items loaded into the knapsack.

[0077] The steps of the spectrum aggregation algorithm are as follows:

[0078] (1) Arrange the bandwidths of the spectrum resources required by N cognitive user services from small to large to obtain a bandwidth set B = {y1, y2, y3, …, yk, …, yN}, where yk is the bandwidth required by cognitive user k's service, yk ≤ yk + 1, k = 1, 2, 3, … N, and B is the bandwidth set required by N cognitive user services.

[0079] (2) Let \(\mathcal{V}\) be the set of idle frequency bands, where [fLi, fHi] represents the i-th idle frequency band, fLi is the low-frequency boundary of idle frequency band i, fHi is the high-frequency boundary of idle frequency band i, and here fHi < fL(i + 1), V + 1 represents the total number of idle frequency bands available for cognitive users within a single spectrum span; if the right boundary of the spectrum span falls within [fL(v + 1), fH(v + 1)], jump to (3); if the right boundary of the spectrum span falls within [fH(v + 1), fL(v + 2)], jump to (5).

[0080] (3) When the right boundary of this spectrum span falls within the idle frequency band [fL(v + 1), fH(v + 1)], remove its bandwidth of xv + 1 from this spectrum span, so that there are V idle frequency bands left within this spectrum span for spectrum aggregation within this spectrum span; <00OO181>[fL(v + 1), fH(v + 1)] can then be used by the next spectrum span, and jump to (4). It should be noted here that the bandwidth of the idle frequency band [fLi, fHi] can be represented by xi, i = 1, 2, …, V + 1.

[0082] (4) In the span, calculate the sum of the bandwidth aggregation of the first n+1 idle frequency bands in sequence for the V idle frequency bands

[0083] Thus we get the set M = {M1, M2, M3, … Mr, …, Mv-1}, n = 1, 2, 3, … v-1, and jump to (6).

[0084] (5) In the span, calculate the sum of the bandwidth aggregation of the first n+1 idle frequency bands in sequence for the V+1 idle frequency bands Thus we get the set M = {M1, M2, M3, … Mr, …, Mv}, and jump to (6).

[0085] (6) Compare each element in M ​​with yk one by one from left to right according to the size of the aggregate bandwidth. When all elements in M ​​are compared, if Mr ≥ yk and there exists an equation Mr = yk, then jump to (7); if Mr ≥ yk and there does not exist an equation Mr = yk, then jump to (8).

[0086] (7) If there is an aggregate bandwidth Mr that satisfies the equation Mr = yk, and a corresponding matching cognitive user k can be found, the idle frequency band corresponding to Mr is allocated to yk to just meet its business needs. In the set B, M, F, the used yk, Mr, [fLi, fHi] are removed to obtain the updated set, and then jump to (9).

[0087] (8) Within the spectrum span, allocate the first aggregated frequency band bandwidth Mr in M ​​that is larger than y1 to the kth cognitive user, that is, the cognitive user k with the bandwidth closest to the aggregated frequency band Mr, to minimize the waste of idle spectrum resources, where k = argmin{yk-Mr}, k = 1, 2, …, N. In the set B, M, F, remove the used yk, Mr, [fLi, fHi] to obtain the updated set and jump to (9).

[0088] (9) If yk is still present in the updated B and Mr is still present in the updated M, and the condition Mr ≥ yk is satisfied, the algorithm is re-executed and jumps to (2); if any of the above conditions is not satisfied, the algorithm ends.

[0089] In summary, the edge computing-based cloud-edge computing power control method and system proposed in this invention, through multi-dimensional technology integration and innovation, has achieved significant breakthroughs in cloud-edge collaborative scheduling, dynamic resource allocation, real-time guarantee, and global energy consumption optimization. It provides an efficient, reliable, and green solution for power systems in complex scenarios and other fields with high real-time requirements. Its core value and innovation are mainly reflected in the following aspects:

[0090] 1. Dynamic Collaboration and Adaptive Scheduling of Cloud-Edge Computing Power

[0091] The core of the present invention is to build a "data-driven + model-driven" dual-engine control system. Through the closed-loop process of steps S1 to S5, the system realizes full-link automation from resource status perception, dynamic decision generation to execution feedback optimization. The introduction of deep reinforcement learning (TD3 algorithm) solves the problem that traditional static strategies are difficult to adapt to dynamic environments: the dual Critic network reduces strategy over-estimation and combines network topology and task priority to generate the optimal allocation plan, which reduces scheduling delays by 15%-20% and increases resource utilization to more than 85%. At the same time, the distributed architecture design (cooperation between the cloud-based main scheduler and the edge sub-scheduler) takes into account both global optimization and local responsiveness, and the task scheduling time is shortened to less than 50ms, which significantly improves the robustness of the system.

[0092] 2. Task Stratification and Enhanced Edge Autonomy

[0093] In response to heterogeneous task requirements, the system implements refined scheduling through a service level agreement (SLA) grading mechanism. Industrial control tasks (such as real-time monitoring of power equipment) are marked as the highest priority and adopt a preemptive scheduling strategy to ensure millisecond-level response; data analysis tasks (such as electricity consumption statistics) dynamically allocate computing power through an elastic resource pool to avoid resource waste. In addition, the edge autonomy module synchronizes with cloud parameters through a lightweight federated learning model, giving edge nodes the ability to make autonomous decisions in an offline environment. For example, in a power system outage scenario, edge nodes can predict load fluctuations based on local data and perform emergency task scheduling to ensure the continuity of critical services (such as grid fault isolation), effectively avoiding the risk of single point failure caused by centralized scheduling relying on the cloud.

[0094] 3. Global Energy Consumption Optimization and Green Computing

[0095] The system innovatively combines genetic algorithms with carbon emission factors to construct a global energy consumption optimization model across cloud-edge nodes. By dynamically adjusting the start and stop states of computing nodes and task allocation weights, it not only balances load and energy costs but also comprehensively considers the real-time electricity prices and clean energy share of the regional power grid to minimize carbon emissions. Experiments have shown that this model can reduce overall system energy consumption by 12%-18% in typical power IoT scenarios, while also reducing carbon emission intensity by over 20%, providing technical support for green computing power scheduling.

[0096] 4. Innovative Breakthroughs in Power Line Communication Technology

[0097] In response to the problems of spectrum resource fragmentation and noise interference in power line carrier communication (PLC), the present invention proposes an OFDM subcarrier allocation optimization method based on quantum genetic algorithm (QGA). By expanding the search space through quantum bit coding, compressing the algorithm complexity by subcarrier clustering, and optimizing the population quality in combination with historical channel data, the system's allocation efficiency in the 917 subcarrier scenario is improved by more than 40%. In addition, the introduction of non-continuous spectrum aggregation technology realizes the dynamic "bundling" of discrete frequency bands through combinatorial optimization theory (such as knapsack problem modeling), and the spectrum utilization rate is increased by 25%-30%, providing high-reliability communication guarantees for scenarios such as new energy microgrids and electric vehicle charging and discharging management.

[0098] 5. Application Prospects and Expanded Value

[0099] The technical framework of the present invention is highly universal and can be widely used in smart grids, industrial Internet of Things, smart cities and other fields. For example, in smart grids, the system can support distributed energy access, real-time load forecasting and dynamic electricity price regulation; in industrial Internet of Things, its edge autonomy can meet the high real-time control requirements of intelligent manufacturing equipment. In the future, combined with 5G network slicing technology and digital twin platforms, the present invention can be further expanded into a cross-industry computing service infrastructure, promoting the construction of an integrated "cloud-edge-end" ecosystem. In addition, by continuously optimizing the federated learning model and quantum computing algorithms, the system's self-learning ability and scheduling accuracy will be further improved, laying the foundation for the research and development of the next generation of edge intelligent systems.

[0100] In summary, through technological innovation and engineering design, this invention has achieved breakthroughs in cloud-edge collaboration, real-time scheduling, energy consumption optimization, and communication efficiency, providing core support for the digital transformation and intelligent upgrade of power systems. This technological achievement not only has significant economic benefits and social value, but also opens up new directions for theoretical research and industrial practice in the field of edge computing.

[0101] In the embodiments disclosed herein, terms such as "installed," "connected," "connected," and "fixed" should be understood broadly. For example, "connected" may refer to a fixed connection, a detachable connection, or an integral connection; and "connected" may refer to a direct connection or an indirect connection via an intermediary. Those skilled in the art will understand the specific meanings of the above terms in the embodiments disclosed herein based on specific circumstances.

[0102] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A cloud-edge computing power control method based on edge computing, characterized in that: The following steps are involved: Step S1: Real-time collection of resource status data of cloud data centers, edge nodes, and terminal devices, including computing power utilization, storage capacity, network bandwidth, task queue length, and energy consumption indicators; Step S2: Build a dynamic scheduling strategy based on the deep reinforcement learning model to generate a computing power allocation plan based on task type, priority, real-time resource status, and historical load data; Step S3: Perform hierarchical processing on tasks, prioritize high-real-time tasks to edge nodes, and dynamically migrate computationally intensive tasks to the cloud or adjacent edge node clusters; Step S4: Through the local autonomous module of the edge node, emergency task scheduling is autonomously performed in the scenario of network disconnection or high latency, and the global resource status of the cloud is synchronously updated; Step S5: Optimize global energy consumption based on genetic algorithms, combine the energy consumption data of cloud and edge nodes, and dynamically adjust the start and stop status of computing nodes and task allocation weights.

2. The cloud-edge computing power control method based on edge computing according to claim 1 is characterized in that: The deep reinforcement learning model in step S2 adopts a double-delay deep deterministic policy gradient TD3 algorithm, whose state space includes node computing power, task delay constraints and network topology relationships, the action space is the task allocation decision vector, and the reward function is composed of task completion time, resource utilization and energy consumption weights.

3. The cloud-edge computing power control method based on edge computing according to claim 1 is characterized in that: The step S3 further includes: dividing the real-time level according to the service level agreement of the task, adopting a preemptive scheduling strategy for industrial control tasks, and adopting a flexible resource reservation mechanism for data analysis tasks.

4. The cloud-edge computing power control method based on edge computing according to claim 1 is characterized in that: In step S4, the local autonomous module has a built-in lightweight decision-making model, which regularly synchronizes parameters with the cloud model through federated learning, supporting edge nodes to perform short-term prediction and task scheduling based on local data in an offline state.

5. A cloud-edge computing power control system based on edge computing according to any one of claims 1-7, characterized in that: include: Resource monitoring unit, used to collect resource status data of the cloud, edge nodes and terminal devices in real time; Dynamic scheduling engine, integrating deep reinforcement learning models to generate and optimize computing power allocation plans; Task hierarchical processor, which prioritizes tasks according to their type and SLA and assigns them to corresponding nodes; The edge autonomy module is deployed on each edge node to support emergency task scheduling in an offline environment; Energy consumption optimization unit, which implements node start and stop strategy with global energy consumption optimization based on genetic algorithm.

6. The cloud-edge computing power control system based on edge computing according to claim 5 is characterized by: The dynamic scheduling engine adopts a distributed architecture, including a cloud-based main scheduler and edge sub-schedulers. The main scheduler is responsible for global resource coordination, and the sub-schedulers quickly respond to task requests based on local status.

7. The cloud-edge computing power control system based on edge computing according to claim 5, characterized in that: The task layered processor supports multi-protocol adaptation, including MQTT, CoAP and custom industrial protocols, and abstracts the computing resources of heterogeneous edge devices through virtualization technology.

8. The cloud-edge computing power control system based on edge computing according to any one of claim 5, characterized in that: The energy consumption optimization unit introduces a carbon emission factor and dynamically adjusts the load distribution of computing nodes based on the real-time electricity price and clean energy proportion of the regional power grid.

Images