Distributed photovoltaic edge virtual cluster coordination control method considering communication time delay

Abstract

The invention relates to a distributed photovoltaic edge virtual cluster coordination control method considering communication time delay, and the method comprises the steps: determining an edge virtual cluster which can carry out task distribution according to the load states of all edge virtual clusters; according to the number of the to-be-allocated tasks and the number of the edge virtual clusters, determining the number N0 of the tasks needing to be allocated for each edge virtual cluster; sequentially distributing the tasks to be distributed to the sorted edge virtual clusters according to a time sequence, wherein N0 tasks to be distributed are distributed to each edge virtual cluster; and in the distribution process of the to-be-distributed task, the weighting and transmission delay between each distributed device and the connected edge virtual cluster is minimized, and the weighting and calculation delay between each edge virtual cluster and the cloud server is minimized. According to the invention, through optimizing the time delay compensation mechanism and the coordination control algorithm, reasonable allocation of computing resources is carried out, and cooperative scheduling between photovoltaic and energy storage is ensured to be more efficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed power systems, and in particular relates to a distributed photovoltaic edge virtual cluster coordination control method taking communication delay into consideration. Background Art

[0002] Under the background of the new power system, distributed photovoltaic power generation is connected to the distribution network on a large scale, and the structure and operation mode of the power grid have changed significantly. As an important component of clean energy, distributed photovoltaic power generation plays an important role in improving the sustainability of the power grid and reducing carbon emissions. However, due to the intermittent and volatile nature of photovoltaic power generation, especially in the case of large photovoltaic power generation on sunny afternoons and low light on rainy days, power quality problems such as voltage exceeding the upper limit are becoming increasingly prominent. In addition, the output of photovoltaic power generation is affected by multiple factors such as seasonal changes and weather conditions, which brings huge challenges to the dispatch of the power system.

[0003] As an effective supplement to photovoltaic power generation, energy storage systems can store excess electricity when photovoltaic power generation is in excess, and release stored electricity when photovoltaic power generation is insufficient, thereby balancing the grid load and improving the stability of the power system. Although energy storage technology has been widely used, the coordinated control between photovoltaics and energy storage still faces many problems, especially in the case of large-scale deployment. How to efficiently and accurately dispatch photovoltaic and energy storage equipment is still a difficult problem that needs to be solved.

[0004] In existing dispatching methods, centralized or distributed control strategies are usually used to coordinate the operation of photovoltaic power generation and energy storage systems. However, most of these methods fail to fully consider the impact of communication delays on system stability and dispatching effects. Since the dispatching between distributed photovoltaic and energy storage devices depends on remote communication, communication delays may cause real-time dispatching response lags, thereby affecting the power balance and power quality of the power grid. In addition, existing control methods often ignore the system's dynamic response capabilities in the face of large-scale photovoltaic access, resulting in low grid dispatching efficiency and difficulty in coping with the instability factors brought about by large-scale photovoltaic grid connection.

[0005] Therefore, how to achieve fast and coordinated control of distributed photovoltaic clusters and energy storage systems while taking into account communication delays has become a key issue in improving grid dispatching efficiency and ensuring power quality. Summary of the invention

[0006] The present invention proposes a distributed photovoltaic edge virtual cluster coordination control method taking communication delay into consideration, thereby solving the technical problems mentioned in the background technology.

[0007] The technical solution of the present invention is as follows: A distributed photovoltaic edge virtual cluster coordination control method considering communication delay, comprising:

[0008] S10: Determine a task allocation edge virtual cluster that can perform task allocation according to the load status of all edge virtual clusters;

[0009] S20: Determine the number of tasks N0 that each task allocation edge virtual cluster needs to allocate according to the number of tasks to be allocated and the number of task allocation edge virtual clusters;

[0010] S30: Arrange the task allocation edge virtual clusters in ascending order according to the computing resource utilization rate, and allocate the tasks to be allocated to the sorted task allocation edge virtual clusters in chronological order, and allocate N0 tasks to be allocated to each task allocation edge virtual cluster;

[0011] S40: In the process of allocating the tasks to be allocated, the weighted sum of transmission delays between each distributed device and the connected edge virtual cluster is minimized, and the weighted sum of calculation delays between each edge virtual cluster and the cloud server is minimized.

[0012] Furthermore, the step S10 includes:

[0013] The resource utilization rates of all edge virtual clusters are obtained. When the resource utilization rate of an edge virtual cluster is lower than a set threshold, the edge virtual cluster is determined to be in a low-load state. The edge virtual cluster in the low-load state is a task allocation edge virtual cluster capable of performing task allocation.

[0014] Further, the resource utilization rate includes: CPU utilization rate, memory utilization rate, storage utilization rate, and the set threshold includes: CPU utilization rate threshold, memory utilization rate threshold, storage utilization rate threshold;

[0015] When the CPU usage of the edge virtual cluster is lower than a CPU usage threshold, the memory usage is lower than a memory usage threshold, and the storage usage is lower than a storage usage threshold, it is determined that the edge virtual cluster is in a low load state.

[0016] Furthermore, the step S30 includes:

[0017] S301: All the tasks to be assigned are organized into a task sequence, and all the tasks to be assigned in the task sequence are arranged in chronological order;

[0018] S302: All the task allocation edge virtual clusters are organized into a cluster sequence, and all the task allocation edge virtual clusters in the cluster sequence are arranged in ascending order according to computing resource usage rates;

[0019] S303: Allocate the first N0 tasks to be assigned in the task sequence to the first task assignment edge virtual cluster in the cluster sequence, and the first N0 tasks to be assigned exit the task sequence;

[0020] S304: Repeat S302 and S303 until all the tasks to be assigned are assigned.

[0021] Furthermore, the step S40 includes:

[0022] Calculate and determine the time slot resources of each distributed device in the edge virtual cluster by using a first optimization formula of the weighted sum transmission delay between each distributed device and the connected edge virtual cluster;

[0023] Through the second optimization formula of weighted and calculated delay between each edge virtual cluster and the cloud server, the task splitting rate, the edge computing resources allocated by the edge virtual cluster to each distributed device, and the cloud computing resources allocated by the cloud server to the distributed devices of each edge virtual cluster are calculated and determined.

[0024] Furthermore, the first optimization formula is:

[0025]

[0026] Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the device transmission delay, τ j,i represents the time slot resource of the i-th distributed device in the j-th edge virtual cluster, L j,i represents the input data size of the i-th distributed device of the j-th edge virtual cluster, R j,i represents the channel capacity of the i-th distributed device of the j-th edge virtual cluster, and T represents the length of a frame.

[0027] Furthermore, the second optimization formula is:

[0028]

[0029] Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the computation delay of the ith distributed device in the jth edge virtual cluster, represents the cloud computing resources allocated by the cloud server to the i-th distributed device of the j-th edge virtual cluster, represents the edge computing resources allocated by the jth edge virtual cluster to the ith distributed device, λ j,i represents the task splitting ratio;

[0030] in, The formula is:

[0031]

[0032] in, represents the computational latency of the edge virtual cluster, represents the transmission delay of the edge virtual cluster, represents the computing delay of the cloud server, L j,i represents the amount of task data, C j,i Indicates the number of CPU cycles required to process each unit of data, W j Denoted as the backhaul communication capacity of each distributed device associated with the jth edge virtual cluster.

[0033] Furthermore, the second optimization formula is optimized and solved by a long-short term neural network model.

[0034] Further, including:

[0035] The time window size, batch size, and number of hidden layer units of the long-term and short-term neural network model are taken as optimization objects;

[0036] The optimization object is optimized by using the optimization algorithm, and the optimal value of the optimization object is used as a hyperparameter to establish a long-term and short-term neural network model;

[0037] The long-term and short-term neural network model is trained, and the second optimization formula is solved by the trained long-term and short-term neural network model.

[0038] Furthermore, the optimization algorithm is a particle swarm algorithm.

[0039] Beneficial effects of the present invention: The present invention optimizes the delay compensation mechanism and the coordinated control algorithm to rationally allocate computing resources, ensure more efficient coordinated scheduling between photovoltaics and energy storage, improve the overall stability and operating efficiency of the power grid, improve the power grid scheduling efficiency and ensure power quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of a distributed photovoltaic edge virtual cluster coordination control method considering communication delay of the present invention.

[0041] Figure 2 It is a structural diagram of the distributed device, edge virtual cluster and cloud server of the present invention.

[0042] Figure 3 Compare communication latency for different architecture configurations. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. The described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0044] In the technical solution of the present invention, Figure 2 As shown, due to the large number of actual control units, there are different time scale scheduling requirements for different types of energy storage equipment and photovoltaic inverters. At the same time, many computing tasks need to be allocated. It is difficult to achieve balanced allocation of cloud server resources while ensuring computing reliability. The scheduling algorithm designed in this patent takes resource utilization as the core and load balancing as the goal. It controls the equipment through the cloud platform and the device layer, and allocates tasks in turn multiple times.

[0045] Figure 1 : is a flowchart provided by a specific process of a distributed photovoltaic edge virtual cluster coordination control method considering communication delay according to the present invention, such as Figure 1 As shown, the present invention includes:

[0046] S10: Determine a task allocation edge virtual cluster that can perform task allocation according to the load status of each of the edge virtual clusters.

[0047] The resource utilization rates of all edge virtual clusters are obtained. When the resource utilization rate of an edge virtual cluster is lower than a set threshold, the edge virtual cluster is determined to be in a low-load state. The edge virtual cluster in the low-load state is a task allocation edge virtual cluster capable of performing task allocation.

[0048] The resource utilization rate includes: CPU utilization rate, memory utilization rate, storage utilization rate, and the set threshold includes: CPU utilization rate threshold, memory utilization rate threshold, storage utilization rate threshold;

[0049] When the CPU usage of the edge virtual cluster is lower than a CPU usage threshold, the memory usage is lower than a memory usage threshold, and the storage usage is lower than a storage usage threshold, it is determined that the edge virtual cluster is in a low load state.

[0050] For the main computing resources related to task processing, namely CPU, memory and storage, certain thresholds are set, and edge virtual clusters that meet the requirements of the following formula are selected. That is, edge virtual clusters whose resource utilization rates are lower than the set thresholds are considered to be in a low-load state and can be used for task allocation. The obtained edge cluster sequence is denoted as D.

[0051] A i =(C i,1 <θ i,1 )∩(C i,2 <θ i,2 )∩(C i,3 <θ i,3 )

[0052] Where: A i is the decision principle of the ith edge virtual cluster; C i,1 is the CPU usage of the ith edge virtual cluster; C i,2 is the memory usage of the ith edge virtual cluster; C i,3 is the storage utilization rate of the i-th edge virtual cluster; θ i,1 is the CPU usage threshold of the i-th edge virtual cluster; θ i,2 is the memory usage threshold of the i-th edge virtual cluster; θ i,3 is the storage usage threshold of the i-th edge virtual cluster.

[0053] S20: Determine the number of tasks N0 that each task allocation edge virtual cluster needs to allocate according to the number of tasks to be allocated and the number of task allocation edge virtual clusters.

[0054]

[0055] Where N t is the number of tasks to be assigned, and N is the number of edge virtual clusters for task assignment.

[0056] S30: Arrange the task allocation edge virtual clusters in ascending order according to the computing resource utilization rate, and allocate the tasks to be allocated to the sorted task allocation edge virtual clusters in chronological order, and allocate N0 tasks to be allocated to each task allocation edge virtual cluster;

[0057] The details are as follows:

[0058] S301: All tasks to be assigned are organized into a task sequence, and all tasks to be assigned in the task sequence are arranged in chronological order.

[0059] The computational analysis tasks to be assigned due to the current scheduling and control system are sorted in order of request time, starting with the earliest requested computation.

[0060] S302: All the task allocation edge virtual clusters are organized into a cluster sequence, and all the task allocation edge virtual clusters in the cluster sequence are arranged in ascending order according to computing resource usage rates.

[0061] The number of task allocation edge virtual clusters contained in the cluster sequence D is N, and the cluster sequence D is sorted. In view of the relatively abundant storage of the server, it is sorted from small to large according to the average utilization rate of CPU and memory to form a cluster sequence D'={S1,S2,…,S N}, where S1, S2, …, S N are the sorted edge virtual clusters. The calculation formula for the average utilization rate is as follows:

[0062]

[0063] In the formula, C i,avr is the average CPU and memory usage of the i-th edge cluster.

[0064] S303: Allocate the first N0 tasks to be assigned in the task sequence to the first task assignment edge virtual cluster in the cluster sequence, and the first N0 tasks to be assigned exit the task sequence.

[0065] Set the task sequence T a The first N0 tasks to be assigned are assigned to the first edge virtual cluster in D'. After completion, the assigned task set is removed from T a Exit.

[0066]

[0067] Where, T1,…,T Nt The tasks to be assigned are sorted by request time.

[0068] S304: Repeat S302 and S303 until all the tasks to be assigned are assigned.

[0069] After the edge cluster allocates computing resources such as CPU and memory to the requested task, return to steps S302 and S303 to reorder the low-load edge clusters and a The remaining task sets in the task sequence are assigned until all tasks in the task sequence are assigned.

[0070] S40: In the process of allocating the tasks to be allocated, the weighted sum of transmission delays between each distributed device and the connected edge virtual cluster is minimized, and the weighted sum of calculation delays between each edge virtual cluster and the cloud server is minimized.

[0071] The following is the specific analysis process: Under the above allocation architecture, energy storage and photovoltaic clusters are divided according to actual changes, which can be regarded as edge terminals. How to reasonably allocate computing tasks between different edge devices and cloud servers to maximize the computing efficiency of the entire system is of great significance for the rapid coordinated control of distributed photovoltaic clusters and energy storage. Next, the coordination problems existing in the allocation process are designed to minimize the computing delay of all tasks by formulating a reasonable computing task allocation strategy.

[0072] In order to minimize computing delays, the computing tasks in the architecture need to be reasonably distributed between the edge and the cloud. As mentioned earlier, in the entire computing process, the segmentation and calculation of tasks require a certain amount of time, which leads to an increase in computing delays. It should be noted that the time required for task segmentation is very small compared to the corresponding computing and communication delays, so it can be ignored. In addition, the calculation results are mostly simple information such as control signals, and their capacity is also very small, so the download delay can be ignored. In summary, there are the following four types of delays in the entire process.

[0073] (1) Transmission delay of devices: Due to limited communication bandwidth, limited computing power, and large computing workload, the overall cloud platform to end-side autonomous control device delay is usually between tens or hundreds of milliseconds. Therefore, the entire execution process of each task will go through multiple time slots. From the perspective of a long time scale, the average transmission delay can be used to represent the average transmission delay of the device. The average transmission delay of the i-th distributed device uploading its computing task to the j-th virtual cluster connected is:

[0074]

[0075] In the formula, T represents the length of a frame, R j,i is the expected channel capacity, which can be achieved by estimating the channel for each device. j,i It represents the normalized value of the number of frames or data that the device needs to transmit in a time frame. Assuming that the time slot resource τ of each device j,i is fixed in each time frame. In this case, the duration of the i-th device sending data is τ j,i , and the other T-τ j,i is its scheduling and waiting time. Therefore, the scheduling and waiting delays are included in the transmission delay.

[0076] This formula expresses that, given a time frame length T, or a normalized magnitude, device i needs to convert L j,i T-level data is sent to the jth edge server, and the channel capacity R that can be maintained is expected to be j,i and duration τ j,iThe average time required to transfer this data.

[0077] (2) Edge virtual cluster computing delay: After receiving all the data from the specific distributed devices at the device layer, the edge virtual cluster server immediately starts the task allocation strategy and divides each task into two parts, one for cloud platform instruction decomposition and the other for edge computing. For general analysis, it is assumed that each computing task can be arbitrarily split. λ∈[0,1] is represented as the task splitting ratio, that is, the proportion of data remaining on the edge server. L j,i That is, the amount of task data that needs to be processed at this time, C j,i is the number of CPU cycles required to process each unit of data. Then, the number of CPU cycles required to process the edge computing part of the data can be expressed as λ j,i L j,i C j,i The computing resources allocated to the i-th device by the j-th edge node are defined as (in CPU cycles / s). Therefore, the computational latency of the edge computing part that processes data at the edge virtual cluster server can be expressed as:

[0078]

[0079] (3) Edge virtual cluster transmission delay: For each edge virtual cluster node, the communication module (transceiver) and the computing module (CPU / GPU) are usually separated. Therefore, the computation of some data in the edge virtual cluster can be parallel to the transmission of some data in the cloud computing. In addition, all edge virtual cluster nodes are connected to the cloud server through different backhaul links, which usually have high bandwidth characteristics. In fact, due to factors such as the randomness of packet arrival, multi-user scheduling, and complex routing algorithms, its latency is difficult to model. Assume that the resource scheduling strategy and routing algorithm are fixed. Therefore, W j is represented as the backhaul communication capacity of each device associated with the jth edge node. Therefore, similar to the average transmission delay, the average backhaul transmission delay is proportional to the size of the data to be offloaded and is expressed as:

[0080]

[0081] In the formula, Indicates the time required for the backhaul link to transmit one bit of data.

[0082] (4) Cloud server computing delay: When the cloud server receives data from the edge virtual cluster node, it allocates the available computing resources to each task and then performs parallel computing. In this way, the number of CPU cycles to process each cloud computing part of the data is (1-λ j,i )L j,i C j,i. Similar to the resource allocation method on the edge virtual cluster server, (in CPU cycles / second) is the cloud computing resource allocated to the i-th device served by the j-th edge virtual cluster node. Therefore, the computational delay for processing the cloud computing part of the data on the cloud server is:

[0083]

[0084] Based on these four types of delays, namely the transmission delay of each device, the computing delay of each edge virtual cluster, the transmission delay of each edge virtual cluster, and the computing delay in the cloud server.

[0085] To determine the total delay of the equipment under the distributed photovoltaic cluster and energy storage fast coordinated control architecture, the following reasonable assumptions are made:

[0086] 1) Since the task splitting operation depends on the specific parameters of each task, such as data size, computing amount, etc., the edge virtual cluster server distributed in each edge virtual cluster node cannot split the task before receiving all the data of the task, so as to better implement the task allocation strategy and ensure the accuracy of the splitting result;

[0087] 2) In actual systems, task calculations may rely on specific data structures and the correlation between adjacent data. For example, for the control of energy storage systems, it is necessary to comprehensively consider information such as the output of distributed power sources and user load requirements. In order to ensure the reliability of the calculation results, the cloud server can only start processing tasks after the transmission between the edge virtual cluster and the cloud server is completed.

[0088] Based on the above assumptions, the total delay of the i-th device of the j-th edge virtual cluster can be expressed as

[0089]

[0090] The optimization goal of the model is to minimize the system delay of all mobile devices. The weighted sum of the total delay of all devices is used as the evaluation criterion, which represents the average delay performance of the entire system. In addition, a positive weight factor β is assigned to each device. j,i ∈[0,1], which represents the fairness among mobile devices and satisfies ∑β j,i = 1. According to the timeliness adjustment priority required by each microgrid device calculation task, that is, determine β j,i To this end, the optimization problem can be expressed as:

[0091]

[0092] T j,i represents the total delay of device i on the jth edge virtual cluster.

[0093] T j,i , τ j,i The following constraints should be met:

[0094]

[0095] In the first constraint, T represents the maximum overall communication resource of all devices, that is, all devices do not exceed the maximum overall communication resource constraint, and the second constraint F c and the third represents the maximum available resources provided by the cloud server and each edge virtual cluster node, that is, the computing resources allocated to the device should not exceed the maximum available resources provided by the cloud server and each edge virtual cluster node, respectively. j,i That is, the task division ratio between nodes.

[0096] The above problem involves four variables that need to be decided. It is a complex problem that is coupled with each other. Therefore, it is necessary to break the problem down into sub-problems and then solve the problem.

[0097] In order to solve the above problem, we first need to analyze the delay components of the computing requirements of each device in the microgrid. From the above delay formula analysis, we can see that the average transmission delay of each device is Only the time slot length τ j,i is determined by the edge virtual cluster node and has nothing to do with other optimization variables. At the same time, the transmission delay from each edge virtual cluster node to the cloud server is The computational latency of each edge virtual cluster node Computational latency in cloud servers Both with τ j,i Taking this into consideration, the distributed photovoltaic cluster and energy storage fast collaborative control architecture can be divided into two independent subsystems, corresponding to the transmission system between the end-side devices and the edge virtual cluster nodes and the collaborative computing system between the edge virtual cluster nodes and the cloud server, thus decomposing the problem into two sub-problems.

[0098] The first sub-problem is to minimize the weighted sum of transmission delays between each end-side device and the connected edge virtual cluster server, that is, to minimize the weighted sum of transmission delays between each distributed device and the connected edge virtual cluster, which can be formulated as:

[0099]

[0100] Communication resource allocation is associated with the weighted sum transmission delay between the devices and the corresponding edge virtual cluster nodes in the distributed photovoltaic cluster and energy storage fast collaborative control architecture. Therefore, the optimal communication resource allocation scheme can be obtained by solving the first sub-problem. For the optimal communication resource allocation strategy of the cloud-edge collaborative system, the time slot allocated to the i-th device served by the j-th edge virtual cluster node can be expressed as:

[0101]

[0102] From the above formula, we can see that the j The optimal time slot of the ith device in an edge virtual cluster is given by the corresponding weight factor β j,i Input data size L j,i and channel capacity R j,i Determine. Weight factor β j,i Determines the importance level of each device in the system, and thus determines the priority of the device. Data size L j,i represents the workload and channel capacity R for sending tasks to the connected microgrid edge servers j,i As the input data increases or the channel conditions deteriorate, the transmission delay of the device will increase accordingly, resulting in the allocation of more time slots.

[0103] The second sub-problem is to minimize the weighted sum calculation delay of each edge virtual cluster node and the cloud server, that is, to minimize the weighted sum calculation delay between each edge virtual cluster and the cloud server, which can be expressed as:

[0104]

[0105] In the formula, The computation delay of the i-th mobile device in the j-th edge virtual cluster can be expressed as:

[0106]

[0107] To this end, the step S40 specifically achieves minimization of the weighted sum calculation delay between each edge virtual cluster and the cloud server and the weighted sum calculation delay between each edge virtual cluster and the cloud server by solving the above two formulas:

[0108] By using a first optimization formula of a weighted sum transmission delay between each distributed device and the connected edge virtual cluster, the time slot resources of each distributed device in the edge virtual cluster and the weighted sum calculation delay between each edge virtual cluster and the cloud server are calculated and determined;

[0109] Through the second optimization formula of weighted and calculated delay between each edge virtual cluster and the cloud server, the task splitting rate, the edge computing resources allocated by the edge virtual cluster to each distributed device, and the cloud computing resources allocated by the cloud server to the distributed devices of each edge virtual cluster are calculated and determined.

[0110] Wherein, the first optimization formula is:

[0111]

[0112] Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the device transmission delay, τ j,i represents the time slot resource of the i-th distributed device in the j-th edge virtual cluster, L j,i represents the input data size of the i-th distributed device of the j-th edge virtual cluster, R j,i represents the channel capacity of the i-th distributed device of the j-th edge virtual cluster, and T represents the length of a frame.

[0113] Wherein, the second optimization formula is:

[0114]

[0115] Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the computation delay of the ith distributed device in the jth edge virtual cluster, represents the cloud computing resources allocated by the cloud server to the i-th distributed device of the j-th edge virtual cluster, represents the edge computing resources allocated by the jth edge virtual cluster to the ith distributed device, λ j,i represents the task splitting ratio;

[0116] in, The formula is:

[0117]

[0118]

[0119] in, represents the computational latency of the edge virtual cluster, represents the transmission delay of the edge virtual cluster, represents the computing delay of the cloud server, L j,i represents the amount of task data, C j,i Indicates the number of CPU cycles required to process each unit of data, W jDenoted as the backhaul communication capacity of each distributed device associated with the jth edge virtual cluster.

[0120] For the second sub-problem, due to its high complexity, an intelligent optimization algorithm can be introduced to find the best solution. Next, the Particle Swarm Optimization (PSO) algorithm is used as an example to construct and solve the fitness function. It should be noted that the PSO algorithm is not the only one that can be selected, and other intelligent optimization algorithms should also be included in the protection scope.

[0121] According to the optimization purpose, the optimization goal is to minimize the weighted and calculated delay of each edge virtual cluster and cloud server. At the same time, the above constraints should be met. Therefore, the fitness function of the PSO optimization strategy can be composed of the weighted and calculated delay of each edge virtual cluster node and cloud server and the penalty function considering the constraints, as shown below:

[0122]

[0123] Where ζ(s) is a strictly increasing sequence, and its form is as follows:

[0124]

[0125] Where s is the number of iterations of the PSO algorithm.

[0126] H1(x), H2(x), and H3(x) are the penalties caused by not being constrained, expressed as follows;

[0127]

[0128] In the formula, θ is a multi-level assignment function, is the level of the penalty function, defined as:

[0129]

[0130] θ and The value of can be determined by the following rules:

[0131]

[0132] Among them, κ is used to measure the degree of violation of the measurement constraint, that is, to express the problem, corresponding to a certain group of τ j,i λ j,i The degree of deviation from the given constraints has an impact on θ and The value of is defined so that the algorithm tends to find more feasible solutions that satisfy the constraints.

[0133] After establishing the fitness function, the PSO algorithm can be used to obtain the optimal result of the variable. The current optimal position of particle i is updated as follows:

[0134]

[0135] The optimal position update formula of the entire particle swarm is as follows.

[0136]

[0137] The velocity and position of each particle in each iteration are updated as follows.

[0138]

[0139] Where i is the number of particles in the swarm, k is the number of iterations, and n and r are random numbers between 0 and 1. c1 and c2 are acceleration factors, which determine the rate of change of the particle's velocity in its own optimal direction and the global optimal direction, respectively. In order to ensure that the particle does not exceed the search space during the iteration process, the following constraints need to be set on the particle's velocity.

[0140]

[0141] Combined with the long short-term memory (LSTM) artificial neural network, LSTM-PSO is used for optimization. The algorithm steps are as follows:

[0142] Step 1: Take the time window size, batch size, and number of hidden layer units in the LSTM model as optimization objects and initialize the adaptive PSO algorithm;

[0143] Step 2: Initialize the size of the particle swarm to m and randomly set the initial position of the particle within the search range. and initial velocity

[0144] Step 3: Pass the position information of each particle to the response behavior model and calculate the fitness function F according to the results predicted by the model. i ,Will Set to the optimal position of the particle swarm The LSTM model is constructed with the corresponding parameters of each particle, trained with the training data, predicted with the validation data, and the mean absolute percentage error of the prediction results is used as the fitness value of each particle.

[0145] Step 4: Determine the global optimal particle position P according to the particle fitness value and population division results best and the local optimal particle position G best , according to the formula Update the position of each particle.

[0146] Step 5: Compare the fitness function values ​​of the optimal positions of each particle The optimal position fitness function value of the particle swarm if Then the optimal position of the particle swarm is The value of The value of

[0147] Step:6: According to the formula:

[0148]

[0149]

[0150] Update the velocity and position of each particle;

[0151] Step 7: Determine the termination condition. If the termination condition is met, return the optimal hyperparameter value; otherwise, continue to loop from step 4, where the termination condition can be reaching the maximum number of iterations;

[0152] Step 8: Build the LSTM model using the optimal hyperparameters;

[0153] Step 9: The model is trained using training data and validation data, and the test set is used for prediction to obtain the prediction results.

[0154] In summary, the present invention uses a long-term and short-term neural network model to optimize and solve the second optimization formula.

[0155] Specifically, they include:

[0156] The time window size, batch size, and number of hidden layer units of the long-term and short-term neural network model are taken as optimization objects;

[0157] The optimization object is optimized by using the optimization algorithm, and the optimal value of the optimization object is used as a hyperparameter to establish a long-term and short-term neural network model;

[0158] The long-term and short-term neural network model is trained, and the second optimization formula is solved by the trained long-term and short-term neural network model.

[0159] Among them, the optimization algorithm is a particle swarm algorithm. It should be noted that the particle swarm algorithm can be selected but not limited to a single one, and other intelligent optimization algorithm processing should also be included in the protection scope.

[0160] In order to verify the effectiveness of the proposed problem model and solution method, a case study was conducted in a certain area in Jiangsu Province. The case study includes 12 end-side devices, corresponding to 9 string photovoltaic inverters and 3 energy storages. The energy storage system is configured with 15kWh, 20kWh, and 20kWh lithium iron phosphate battery packs, and the photovoltaic configuration is 6 groups of 10kWp photovoltaic modules and 3 groups of 15kWp photovoltaic modules. The end-side devices are divided into clusters according to the output, load demand, and energy storage status of each group of string inverters, and the data is uploaded to the cloud platform. The cloud platform reasonably allocates resources, reduces communication latency, and realizes rapid coordinated control of distributed photovoltaics and energy storage.

[0161] According to the historical operation data of the substation, the terminal devices are divided into three clusters, named et1, et2, and et3. The following typical scenarios are set for verification.

[0162] Table 1 Typical scenario simulation examples in substation areas

[0163]

[0164] The communication delay after cloud platform resource configuration is shown in the following table.

[0165] Table 2 Resource configuration communication delay summary

[0166]

[0167] According to different typical scenarios, in general, when the proportion of photovoltaic output is high (such as 12 noon on a sunny day), due to the large amount of power generated by the photovoltaic system and the large amount of data, more data needs to be processed during calculation and analysis, so the communication delay increases slightly, the data processing load is heavy, and the total delay increases accordingly (such as 0.9171 seconds). In the period when the photovoltaic output is low (such as in the morning or rainy weather), the photovoltaic system generates less power and processes less data, so the communication delay is low and the total delay is also reduced (such as 0.7198 seconds). In addition, weather conditions also have an impact on the delay: sunny and cloudy weather usually leads to higher photovoltaic output, which increases the complexity of data calculation and causes a slight increase in communication delay; while rainy weather reduces photovoltaic output, thereby reducing communication delay. In general, the communication delay fluctuates slightly based on typical scenario 1, and this fluctuation is within a reasonable range, which is in line with the expectations in actual operation, ensuring that the system can effectively respond to changes in communication delay under different scenarios.

[0168] The following different conditions are set for simulation verification, considering the communication delay resource configuration architecture and the traditional centralized configuration architecture, and comparing the communication delay.

[0169] like Figure 3As shown, the simulation results prove that the corresponding constraints are guaranteed, indicating that the proposed problem model and solution method meet the constraints requirements. Therefore, using the proposed method to solve the proposed problem model can obtain a business cloud-edge resource configuration solution that meets the requirements.

[0170] The communication delays under the two architectures are significantly different. The communication delay of the traditional centralized configuration architecture is generally higher than that of the cloud platform resource configuration architecture, especially during periods of high photovoltaic output (such as 12:00 on sunny days). The traditional architecture has a significantly higher delay due to the need to centrally process a large amount of data, while the cloud platform architecture has a smaller increase in data delay when processing high output, showing strong adaptability. When the photovoltaic output is low, the delay difference between the two is small. In general, the cloud platform architecture has lower delay when photovoltaic power generation is high, especially during high output periods such as noon, and can provide more efficient communication support. The average communication delay of the traditional architecture is 1.5639 seconds, while that of the cloud platform architecture is 0.78935 seconds, with a significant gap. It can be seen that the distributed photovoltaic edge virtual cluster coordination control method considering communication delay proposed in the present invention has obvious advantages in communication efficiency and delay control, and can more effectively support real-time scheduling and grid stability.

[0171] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A distributed photovoltaic edge virtual cluster coordination control method considering communication delay, characterized in that: include: S10: Determine a task allocation edge virtual cluster that can perform task allocation according to the load status of all edge virtual clusters; S20: Determine the number of tasks N0 that each task allocation edge virtual cluster needs to allocate according to the number of tasks to be allocated and the number of task allocation edge virtual clusters; S30: Arrange the task allocation edge virtual clusters in ascending order according to the computing resource utilization rate, and allocate the tasks to be allocated to the sorted task allocation edge virtual clusters in chronological order, and allocate N0 tasks to be allocated to each task allocation edge virtual cluster; S40: In the process of allocating the tasks to be allocated, the weighted sum of transmission delays between each distributed device and the connected edge virtual cluster is minimized, and the weighted sum of calculation delays between each edge virtual cluster and the cloud server is minimized.

2. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 1 is characterized in that: The step S10 comprises: The resource utilization rates of all edge virtual clusters are obtained. When the resource utilization rate of an edge virtual cluster is lower than a set threshold, the edge virtual cluster is determined to be in a low-load state. The edge virtual cluster in the low-load state is a task allocation edge virtual cluster capable of performing task allocation.

3. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 2 is characterized in that: The resource utilization rate includes: CPU utilization rate, memory utilization rate, storage utilization rate, and the set threshold includes: CPU utilization rate threshold, memory utilization rate threshold, storage utilization rate threshold; When the CPU usage of the edge virtual cluster is lower than a CPU usage threshold, the memory usage is lower than a memory usage threshold, and the storage usage is lower than a storage usage threshold, it is determined that the edge virtual cluster is in a low load state.

4. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 1, characterized in that: The step S30 comprises: S301: All the tasks to be assigned are organized into a task sequence, and all the tasks to be assigned in the task sequence are arranged in chronological order; S302: All the task allocation edge virtual clusters are organized into a cluster sequence, and all the task allocation edge virtual clusters in the cluster sequence are arranged in ascending order according to computing resource usage rates; S303: Allocate the first N0 tasks to be assigned in the task sequence to the first task assignment edge virtual cluster in the cluster sequence, and the first N0 tasks to be assigned exit the task sequence; S304: Repeat S302 and S303 until all the tasks to be assigned are assigned.

5. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 1, characterized in that: The step S40 comprises: Calculate and determine the time slot resources of each distributed device in the edge virtual cluster by using a first optimization formula of the weighted sum transmission delay between each distributed device and the connected edge virtual cluster; Through the second optimization formula of weighted and calculated delay between each edge virtual cluster and the cloud server, the task splitting rate, the edge computing resources allocated by the edge virtual cluster to each distributed device, and the cloud computing resources allocated by the cloud server to the distributed devices of each edge virtual cluster are calculated and determined.

6. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 5, characterized in that: The first optimization formula is: Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the device transmission delay, τ j,i represents the time slot resource of the i-th distributed device in the j-th edge virtual cluster, L j,i represents the input data size of the i-th distributed device of the j-th edge virtual cluster, R j,i represents the channel capacity of the i-th distributed device of the j-th edge virtual cluster, and T represents the length of a frame.

7. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 6, characterized in that: The second optimization formula is: Among them, β j,i represents the weight factor of the i-th distributed device in the j-th edge virtual cluster, represents the computation delay of the ith distributed device in the jth edge virtual cluster, represents the cloud computing resources allocated by the cloud server to the i-th distributed device of the j-th edge virtual cluster, represents the edge computing resources allocated by the jth edge virtual cluster to the ith distributed device, λ j,i represents the task splitting ratio; in, The formula is: in, represents the computational latency of the edge virtual cluster, represents the transmission delay of the edge virtual cluster, represents the computing delay of the cloud server, L j,i represents the amount of task data, C j,i Indicates the number of CPU cycles required to process each unit of data, W j Denoted as the backhaul communication capacity of each distributed device associated with the jth edge virtual cluster.

8. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 7, characterized in that: The second optimization formula is solved by using a long-term and short-term neural network model.

9. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 8, characterized in that: include: The time window size, batch size, and number of hidden layer units of the long-term and short-term neural network model are taken as optimization objects; The optimization object is optimized by using the optimization algorithm, and the optimal value of the optimization object is used as a hyperparameter to establish a long-term and short-term neural network model; The long-term and short-term neural network model is trained, and the second optimization formula is solved by the trained long-term and short-term neural network model.

10. The distributed photovoltaic edge virtual cluster coordination control method considering communication delay as claimed in claim 9, characterized in that: The optimization algorithm is a particle swarm optimization algorithm.

Images