Power distribution network data high-frequency acquisition and low-delay interaction method and system

Abstract

The invention discloses a power distribution network data high-frequency acquisition and low-delay interaction method and system, and the method comprises the steps: firstly, constructing a communication cooperative network model, taking the minimum end-to-end delay of power distribution service data and the weighted sum of network operation cost as a target, converting an optimization problem into a form suitable for real-time solving through employing a Lyapunov optimization method, and carrying out the real-time solving of the optimization problem; and further based on the optimization problem, constructing a Markov model, improving a loss function of a deep Q network (DQN) to adapt to the acquisition requirements of distribution network services, and designing a corresponding network learning training method to optimize terminal scheduling and data compression decisions. And finally, a transmission power and bandwidth distributed optimization method based on an alternating direction multiplier method (ADMM) is adopted, so that the efficiency of network resource allocation is further improved. According to the technical scheme, an optimal data acquisition strategy with low time delay, low cost and balanced scheduling is realized, and the method is suitable for high-frequency data acquisition scenes such as a smart power grid and an industrial internet of things.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and specifically to a method and system for high-frequency collection and low-latency interaction of distribution network data. Background Art

[0002] With the development of smart grid technology and the large-scale access of new energy sources, the operation status of distribution networks has become more complex and dynamic. The monitoring, maintenance and optimization of modern distribution networks require real-time and accurate data support, which puts higher requirements on the frequency of data collection and the real-time transmission. Network resource allocation is one of the main means to meet this requirement. However, the current resource allocation method is often based on isolated optimization of one side of the communication side or the perception side. This isolated optimization is difficult to fully utilize resources, and the data transmission efficiency is low, which seriously affects the real-time monitoring business and fault response capabilities of the power grid. It is easy to have performance bottlenecks and cannot meet the stringent requirements of high-frequency collection business for real-time response capabilities.

[0003] The collaborative allocation of synaesthesia resources brings a new solution to the high-frequency data collection business of power distribution. By deeply integrating communication and perception resources and jointly optimizing perception data compression, transmission bandwidth allocation and power control, synaesthesia resources can be allocated and utilized more effectively, improving the overall performance of the system. The collaborative allocation of synaesthesia resources also needs to solve the following technical challenges:

[0004] First, there is a contradiction between latency and network operating costs. How to ensure low-latency interaction of distribution network services while avoiding a surge in network operating costs is one of the technical challenges of the present invention. In addition, the coordinated allocation of highly heterogeneous multi-dimensional resources is a mixed nonlinear integer programming problem with high optimization complexity.

[0005] Second, although machine learning can provide a model-free solution to complex optimization problems, traditional methods lack awareness of scheduling frequency during the strategy learning process and may violate the constraints of the optimization problem in exchange for maximizing their own profits, which cannot meet the data collection needs of distribution network services. Summary of the invention

[0006] In order to solve the technical problems existing in the background technology, the present invention aims to provide a method and system for high-frequency collection and low-latency interaction of distribution network data.

[0007] In order to solve the technical problem, the technical solution of the present invention is:

[0008] A method for high-frequency collection and low-latency interaction of distribution network data based on synaesthesia collaboration, the method comprising:

[0009] Construct a synergetic collaborative network model, with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the operation cost of the synergetic collaborative network, construct an optimization problem, apply the Lyapunov optimization method, transform the constructed optimization problem into a form suitable for solving, and jointly optimize dynamic scheduling, resource allocation and power control;

[0010] Based on the optimization problem, a Markov model is constructed, and the DQN loss function is improved to adapt to the distribution network service collection needs. The DQN network learning and training method is designed to optimize terminal scheduling and data compression decisions. Finally, through the transmission power and bandwidth distributed optimization method based on the ADMM algorithm, the network resource allocation is further optimized to achieve the optimal data collection strategy with low latency, low cost and balanced scheduling.

[0011] Furthermore, the synaesthesia collaborative network model includes: a collection terminal scheduling model, a perception data compression and transmission model, an edge data processing model, and an end-to-end delay and network operation cost model;

[0012] The total optimization time is divided into T time slots, and the set is expressed as The length of each time slot is τ 0 ,Each time slot includes three processes: perception data collection, data compression and transmission, and edge data processing;

[0013] Among them, building a collection terminal scheduling model includes:

[0014] Determine the number of terminals at the terminal layer as J, assuming that the gateway schedules I in each time slot th <J terminals collect and upload data, and determine the terminal scheduling indicator variable as x j (t), x j (t) = 1 indicates terminal d j Scheduled in the tth time slot, x j (t) = 0 means not scheduled, and the terminal d of the tth time slot is determined. j The amount of sensory data collected is a j (t), the amount of waiting compression is expressed as:

[0015] Z j (t) = x j (t)a j (t) (1)

[0016] Among them, the construction of perception data compression and transmission model includes:

[0017] Considering the redundancy of the original data, the acquisition terminal needs to compress the data to determine d j The data compression rate is It is related to the data compression ratio y j (t) is related to, expressed as:

[0018]

[0019] In the formula, is the compression complexity, that is, the number of CPU cycles required to compress each bit of data, is d j Available CPU frequency, d j Data compression delay And data compression energy consumption Respectively expressed as:

[0020]

[0021]

[0022] In the formula, ζ j is d j Energy consumption coefficient;

[0023] During the data compression process, the data is transmitted to the gateway synchronously and in parallel. j Data transmission rate R between the gateway j (t) is expressed as:

[0024]

[0025] In the formula, B j (t) is d j The transmission bandwidth, h j (t) is the channel gain, e j (t) and ζ 0 are the electromagnetic interference power and Gaussian white noise power during the transmission process, d j Data transmission delay and transmission energy consumption Respectively expressed as:

[0026]

[0027] In the formula, A j (t) = Z j (t)y j (t) is d j The amount of data transmitted to the gateway, that is, the amount of data arriving at the edge queue;

[0028] Among them, building an edge data processing model includes:

[0029] The gateway maintains a data queue for each terminal to store undecompressed data. j For example, the input of the queue in the tth time slot is the amount of data A transmitted to the gateway. j(t), the output of the tth time slot is the amount of data U processed by edge decompression j (t), therefore, d j The data queue evolution formula is:

[0030] Q j (t+1)=Q j (t)-U j (t)+A j (t) (8)

[0031]

[0032] In the formula, The number of CPU cycles required to decompress each bit of data, Decompress and process for the gateway j CPU frequency of queue data;

[0033] Based on Little's law, the edge side d j Data queuing delay of the queue It is expressed as:

[0034]

[0035] In the formula, Q j The average data arrival rate (t) is expressed as:

[0036]

[0037] The end-to-end delay and network operation cost model is constructed, including:

[0038] In the synaesthesia synergy network, d j The end-to-end delay is composed of the data compression delay Data transmission delay Data queuing delay In which data compression and data transmission are parallel processes, and the end-to-end delay should be taken into account. and The larger value between It is expressed as:

[0039]

[0040] The cost of running a collaborative network Data compression energy consumption and data transmission energy consumption Related, expressed as:

[0041]

[0042] In the formula, ξj is the cost coefficient.

[0043] Furthermore, the construction optimization problem specifically includes:

[0044] In order to maintain the operating level of the distribution business carried by each acquisition terminal, the terminal must complete a certain number of data acquisition, compression and transmission within the specified time period; the total optimization time is further divided into I time periods, each of which consists of T 0 It is composed of consecutive time slots, that is, T = IT 0 , the relationship between the i-th period and the time slot is The terminal scheduling frequency constraint is:

[0045]

[0046] Where N j is d j The lower limit of the number of dispatches;

[0047] Aiming at the low-latency interactive demand for high-frequency data collection in the distribution network, the weighted sum of the end-to-end delay of the distribution business data and the operation cost of the synergistic collaborative network is minimized through synergistic collaborative optimization, that is, joint optimization of terminal scheduling, data compression ratio, bandwidth allocation and power control. The optimization problem is expressed as:

[0048]

[0049] In the formula, α is the weight of the network operation cost, which is used to balance the order of magnitude and achieve the optimization target trade-off, B max is the total available bandwidth for data transmission, P min and P max are the upper and lower limits of the transmission power, y j,max and j,min They are d j The upper and lower limits of the compression ratio, C 1 and C 2 is the terminal scheduling indicator variable constraint, C 3 is the transmission power constraint, C 4 is the bandwidth allocation constraint, C 5 Select a constraint for the compression ratio, is the long-term terminal scheduling frequency virtual queue, and V represents the weight of the virtual queue, which is used to achieve a trade-off between minimizing queue drift and minimizing the weighted sum of end-to-end delay and network operation cost.

[0050] Furthermore, the construction of the Markov model specifically includes:

[0051] P1 is modeled as a Markov decision MDP model, and the agent maintained by the gateway is composed of an evaluation network Target Network As well as the experience replay pool, the MDP model includes three elements: state space, action space and reward, as follows:

[0052] State space: The state space is defined as the set of terminal states, expressed as Terminal d j Status S j (t) includes the amount of perceived data, available CPU frequency, compression complexity, edge queue backlog, terminal scheduling frequency virtual queue deficit, and is expressed as:

[0053]

[0054] Action space: Define the action space as the terminal scheduling decision x j (t) and data compression ratio decision y j (t) composed of a set It is expressed as:

[0055] M j (t) = {x j (t),y j (t)} (17)

[0056] Reward: Define reward θ(t) as the opposite of the optimization objective, expressed as:

[0057]

[0058] Furthermore, the improved DQN loss function specifically includes:

[0059] The DQN loss function reflects the learning performance of the intelligent agent. By introducing a penalty term to measure the satisfaction of the long-term terminal scheduling frequency constraint in the loss function calculation, the algorithm's perception of the scheduling frequency constraint is improved. Specifically, when the constraint satisfaction is worse, the penalty term is larger, indicating that the actual benefit obtained under the current action is lower than the estimate. The proposed algorithm will constrain such actions and encourage the DQN network to explore or develop other better actions, so as to achieve more reasonable synaesthesia collaborative terminal scheduling and data compression decisions. The loss function including the penalty term is calculated as:

[0060]

[0061] In the formula, 1{t=iT 0} is an indicator function, if and only if t = iT 0 The value is 1 when γ γ is the discount factor, β(t) is the experience data set sampled from the experience replay pool, is the penalty factor, based on the reward, per T 0If the terminal scheduling frequency constraint is not met in a time slot, a penalty item needs to be subtracted from the original reward.

[0062] Furthermore, the DQN network learning specifically includes:

[0063] First, at the beginning of each time slot, the state space is input into the evaluation network In the example above, we can obtain the Q value of taking different actions. And select the action with the largest Q value, which is:

[0064]

[0065] The agent performs action M * (t);

[0066] Then, at the end of each time slot, the agent observes the end-to-end delay, network operation cost, and virtual queue deficit performance, calculates the reward θ(t) according to equation (18), and updates Q j (t) and And transfer the state to S(t+1), then, the experience data [S(t),M * (t),θ(t),S(t+1)] are stored in the experience replay pool;

[0067] Finally, the agent calculates the loss function based on formula (19) and updates the evaluation network according to the gradient descent method. Every T 0 The time slot will target the network and Alignment.

[0068] Furthermore, the transmission power and bandwidth distributed optimization based on the ADMM algorithm specifically includes:

[0069] Based on the known terminal scheduling strategy and data compression ratio, the optimization problem P1 is further transformed into:

[0070]

[0071] sC 4 ~C 5 (twenty one)

[0072] Step 1: To simplify the original constraints, decompose the problem into a form that is easy to solve by Lagrange multipliers and introduce auxiliary variables w j =B j (t) and The optimization problem is transformed into:

[0073]

[0074] Step 2: Establish the augmented Lagrangian function of P3, expressed as:

[0075]

[0076] In the formula, λ j , μ j is the Lagrange multiplier, ρ is the penalty factor, is the L2 norm, are the vectors corresponding to each variable set;

[0077] Select Initial Value

[0078] Step 3: Update the auxiliary variables, the transmission power of each terminal, the bandwidth and the Lagrange multiplier, which can be expressed as:

[0079]

[0080] P min ≤g j ≤P max (26)

[0081]

[0082] Where n is the number of iterations;

[0083] Step 4: The number of iterations increases by 1;

[0084] Step 5: Determine the convergence of the algorithm. The algorithm terminates if the following conditions are met:

[0085]

[0086] In the formula, ε 1 , ε 2 is the preset tolerance;

[0087] Finally, the gateway will converge to the transmission power obtained in the last iteration. and bandwidth as the allocation decision for the current time slot.

[0088] A distribution network data high-frequency collection and low-latency interactive system based on synaesthesia collaboration, the system comprising: a device layer, a perception layer, an edge layer and a master station layer;

[0089] The equipment layer: includes distributed photovoltaics, distributed wind turbines, smart charging piles, energy storage batteries and various electrical equipment such as power distribution equipment;

[0090] The perception layer includes a plurality of high-frequency distribution network data collection terminal devices, which are used to perceive the key data of electrical quantity, state quantity and network environment quantity during the operation of various electrical equipment, and compress and package the perceived data according to the received scheduling and compression scheme, and upload the compressed data to the edge layer according to the received bandwidth and transmission power allocation strategy;

[0091] The edge layer: includes a high-frequency data collection and low-latency interactive gateway device for distribution network based on synaesthesia collaboration and a 5G base station; the 5G base station provides communication coverage for the perception layer, and the high-frequency data collection and low-latency interactive gateway device for distribution network based on synaesthesia collaboration is responsible for decompressing and processing the uploaded compressed data, and uploading it to the main station layer through optical fiber, providing low-latency data support for the operation of distribution business, and issuing strategies for the perception layer;

[0092] The master station layer includes a power grid control center, which is responsible for supporting cross-regional coordinated control of distribution networks, panoramic monitoring of distribution networks, and main and distribution coordinated control business operations.

[0093] Furthermore, the distribution network data high-frequency collection terminal device of the perception layer in the synaesthesia-coordinated distribution network data high-frequency collection and low-latency interactive system includes: a collection module, a compression module, a scheduling perception module and a communication module;

[0094] The acquisition module includes a high-frequency data acquisition terminal for distribution network deployed on various electrical equipment, which is responsible for sensing the key data of electrical quantity, state quantity and network environment quantity during the operation of various electrical equipment;

[0095] The compression module is responsible for compressing and packaging the original data according to the instruction of the scheduling perception module;

[0096] The scheduling perception module is responsible for determining whether to upload data according to the decision issued by the edge layer;

[0097] The communication module is responsible for transmitting data through the 5G channel, receiving the scheduling and compression schemes and bandwidth and transmission power allocation decisions fed back by the edge layer, and sending compressed and packaged key equipment operation data to the edge layer.

[0098] Furthermore, the edge layer's synaesthesia-based high-frequency data collection and low-latency interactive gateway device for distribution networks includes: a synaesthesia-based model building module, an optimization problem building and transformation module, a Q network learning module, a scheduling frequency-aware loss function improvement module, an action selection module, an augmented Lagrangian function building module, a distributed optimization iteration module, and a communication module.

[0099] The synaesthesia collaboration model building module is responsible for building the acquisition terminal scheduling model, the perception data compression and transmission model, the edge data processing model, and the end-to-end delay and network operation cost model;

[0100] The optimization problem construction and transformation module is responsible for constructing the weighted sum optimization problem of minimizing the end-to-end delay of distribution business data and the operation cost of the synergy collaborative network based on the synergy collaborative model, and transforming the problem based on Lyapunov;

[0101] The Q network learning module is responsible for maintaining a set of DQN networks and experience replay pools for each terminal;

[0102] The scheduling frequency-aware loss function improvement module is responsible for introducing a penalty term reflecting the terminal acquisition frequency to improve the calculation process of the loss function;

[0103] The action selection module is responsible for selecting the action with the largest Q value based on the DQN network as the terminal scheduling and data compression ratio selection decision;

[0104] The augmented Lagrangian function construction module is responsible for introducing auxiliary variables and constructing an augmented Lagrangian function for the joint optimization problem of transmission power and bandwidth;

[0105] The distributed optimization iteration module is responsible for updating auxiliary variables, transmission power of each terminal, bandwidth and Lagrange multipliers until the algorithm converges;

[0106] The communication module is responsible for communicating with the 5G base station, issuing terminal scheduling, data compression ratio scheme, and bandwidth and transmission power allocation decisions, receiving data uploaded by the high-frequency acquisition terminal, and uploading data to the main station.

[0107] A computer device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the above-mentioned methods and systems for high-frequency collection and low-latency interaction of distribution network data is implemented.

[0108] A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the program is executed by a processor, it implements any one of the above-mentioned methods and systems for high-frequency collection and low-latency interaction of distribution network data.

[0109] Compared with the prior art, the advantages of the present invention are:

[0110] (1) The present invention proposes a synaesthesia collaborative intelligent resource allocation algorithm for high-frequency acquisition scheduling frequency perception. By gradually processing in stages, it can effectively decouple the synaesthesia resource collaborative allocation problem and realize the collaborative optimization of acquisition terminal scheduling and data compression ratio optimization and transmission power control and bandwidth allocation. A penalty item that reflects the satisfaction of long-term terminal scheduling frequency constraints is set to improve the scheduling frequency perception ability in the strategy learning process, avoid violating the problem constraints in exchange for maximizing benefits, and has good applicability to meet the high-frequency acquisition needs of new energy services in the distribution network.

[0111] (2) The present invention proposes a high-frequency data collection and low-latency interactive system for distribution network based on synaesthesia, including a device layer, a perception layer, an edge layer and a master station layer. The device layer includes various types of electrical equipment such as distributed photovoltaics, distributed wind turbines, smart charging piles, energy storage batteries, and distribution equipment. The perception layer is composed of multiple high-frequency data collection terminals for distribution network, which can perceive the key data in the operation process of various types of electrical equipment, and compress and package them and upload them to the edge layer. The edge layer is composed of a high-frequency data collection and low-latency interactive gateway for distribution network based on synaesthesia and a 5G base station. The base station provides communication coverage for the perception layer. The high-frequency data collection and low-latency interactive gateway for distribution network based on synaesthesia is not only responsible for decompressing and processing the uploaded compressed data, and uploading it to the master station layer through optical fiber, but also needs to issue decisions for the perception layer. The master station layer includes a power grid control center, which is responsible for supporting business operations such as cross-regional collaborative control of distribution networks, panoramic monitoring of distribution networks, and collaborative control of main and distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1 , distribution network data high-frequency collection and low-latency interactive system diagram based on synaesthesia collaboration;

[0113] Figure 2 , the composition diagram of the distribution network data high-frequency collection terminal device;

[0114] Figure 3 , the composition diagram of the high-frequency collection of distribution network data and low-latency interactive gateway device based on synaesthesia collaboration;

[0115] Figure 4 , flow chart of the high-frequency collection and low-latency interactive method of distribution network data based on synaesthesia collaboration. DETAILED DESCRIPTION

[0116] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0117] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0118] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0119] Embodiment 1:

[0120] This embodiment proposes a high-frequency data collection and low-latency interactive system for distribution network based on synaesthesia collaboration, including a device layer, a perception layer, an edge layer and a master station layer. Figure 1 shown.

[0121] Equipment layer: includes distributed photovoltaics, distributed wind turbines, smart charging piles, energy storage batteries, power distribution equipment and other electrical equipment.

[0122] Perception layer: It includes multiple high-frequency distribution network data collection terminal devices, which can perceive key data such as electrical quantity, status quantity, network environment quantity, etc. during the operation of various electrical equipment, and compress and package the perceived data according to the received scheduling and compression scheme, and upload the compressed data to the edge layer according to the received bandwidth and transmission power allocation strategy.

[0123] Edge layer: It is composed of a high-frequency data collection and low-latency interactive gateway device for distribution network based on synergy and a 5G base station. The 5G base station provides communication coverage for the perception layer. The high-frequency data collection and low-latency interactive gateway device for distribution network based on synergy is not only responsible for decompressing and processing the uploaded compressed data, and uploading it to the main station layer through optical fiber, providing low-latency data support for the operation of distribution business, but also needs to issue strategies to the perception layer.

[0124] Main station layer: includes the power grid control center, which is responsible for supporting cross-regional coordinated control of distribution networks, panoramic monitoring of distribution networks, main and distribution coordinated control and other business operations.

[0125] The high-frequency data collection terminal device of the distribution network based on synaesthesia and low-latency interactive system consists of a collection module, a compression module, a scheduling perception module and a communication module. Figure 2 As shown in the figure, each functional module is introduced as follows:

[0126] Collection module: includes distribution network data high-frequency collection terminals deployed on various electrical equipment, responsible for sensing key data such as electrical quantity, state quantity, network environment quantity, etc. during the operation of various electrical equipment;

[0127] Compression module: responsible for compressing and packaging the original data according to the instructions of the scheduling perception module;

[0128] Scheduling perception module: responsible for determining whether to upload data based on the decision issued by the edge layer;

[0129] Communication module: responsible for transmitting data through 5G channels, receiving scheduling and compression schemes as well as bandwidth and transmission power allocation decisions from the edge layer, and sending compressed and packaged key equipment operation data to the edge layer.

[0130] The high-frequency data collection and low-latency interactive gateway device based on synaesthesia collaboration at the edge layer of the system consists of a synaesthesia collaboration model construction module, an optimization problem construction and transformation module, a Q network learning module, a scheduling frequency-aware loss function improvement module, an action selection module, an augmented Lagrangian function construction module, a distributed optimization iteration module, and a communication module. Figure 3 As shown in the figure, each functional module is introduced as follows:

[0131] Synaesthesia collaboration model building module: responsible for building the acquisition terminal scheduling model, perception data compression and transmission model, edge data processing model, and end-to-end delay and network operation cost model;

[0132] Optimization problem construction and transformation module: responsible for constructing the weighted sum optimization problem of minimizing the end-to-end delay of distribution business data and the operation cost of the synergistic collaborative network based on the synergistic collaborative model, and transforming the problem based on Lyapunov;

[0133] Q network learning module: responsible for maintaining a set of DQN networks and experience replay pools for each terminal;

[0134] Scheduling frequency-aware loss function improvement module: responsible for introducing a penalty term that reflects the terminal collection frequency to improve the calculation process of the loss function;

[0135] Action selection module: responsible for selecting the action with the largest Q value based on the DQN network as the terminal scheduling and data compression ratio selection decision;

[0136] Augmented Lagrangian function construction module: responsible for introducing auxiliary variables and constructing the augmented Lagrangian function for the joint optimization problem of transmission power and bandwidth;

[0137] Distributed optimization iteration module: responsible for updating auxiliary variables, transmission power of each terminal, bandwidth and Lagrange multipliers until the algorithm converges;

[0138] Communication module: responsible for communicating with 5G base stations, issuing terminal scheduling, data compression ratio solutions, and bandwidth and transmission power allocation decisions, receiving data uploaded by high-frequency acquisition terminals, and uploading data to the main station.

[0139] Embodiment 2:

[0140] The present invention proposes a method for high-frequency collection and low-latency interaction of distribution network data based on synaesthesia collaboration, which mainly includes a method for constructing a network model for high-frequency collection and low-latency interaction of distribution network data based on synaesthesia collaboration and a synaesthesia collaboration intelligent resource allocation algorithm for high-frequency collection scheduling frequency perception. The specific process is as follows: Figure 4 shown.

[0141] 1. Synaesthesia-based high-frequency data collection and low-latency interactive network model construction method

[0142] The present invention proposes a method for constructing a synergistic distribution network data high-frequency collection and low-latency interactive network model. First, a synergistic collaborative network model is proposed; secondly, an optimization problem is constructed with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the synergistic collaborative network operation cost, and the optimization problem is transformed based on Lyapunov.

[0143] (1) Synaesthesia collaborative network model

[0144] The present invention divides the total optimization time into T time slots, and the set is expressed as The length of each time slot is τ 0 . Each time slot includes three processes: perception data collection, data compression and transmission, and edge data processing. In the process of perception data collection, considering that a large number of terminals collecting data at the same time will lead to data congestion and edge data backlog, the gateway makes terminal scheduling decisions, and the scheduled terminals collect perception data in real time; in the process of data compression and transmission, the perception data contains a lot of redundant information. In order to improve the efficiency of edge layer data processing and reduce the end-to-end data delay and the operating cost of the synergetic collaborative network, the gateway makes data compression ratio, bandwidth allocation, and transmission power control decisions for the terminal. The terminal compresses and transmits the perception data to the gateway for processing based on the decision.

[0145] 1) Collection terminal scheduling model

[0146] Define the number of terminals in the terminal layer as J. Assume that each time slot gateway can schedule I th <J terminals collect and upload data. Define the terminal scheduling indicator variable as x j (t), x j (t) = 1 indicates terminal d j Scheduled in the tth time slot, x j (t) = 0 means not scheduled. Define the tth time slot terminal d jThe amount of sensory data collected is a j (t), the amount of waiting compression is expressed as

[0147] Z j (t) = x j (t)a j (t) (1)

[0148] 2) Perception data compression and transmission model

[0149] Considering the redundancy of the original data, the acquisition terminal needs to compress the data. j The data compression rate is It is related to the data compression ratio y j (t) is related to

[0150]

[0151] In the formula, is the compression complexity, that is, the number of CPU cycles required to compress each bit of data. is d j Available CPU frequencies. j Data compression delay And data compression energy consumption Respectively expressed as

[0152]

[0153]

[0154] In the formula, ζ j is d j The energy consumption coefficient.

[0155] During the data compression process, the data is transmitted to the gateway synchronously and in parallel. j Data transmission rate R between the gateway j (t) is expressed as

[0156]

[0157] In the formula, B j (t) is d j The transmission bandwidth, h j (t) is the channel gain, e j (t) and ζ 0 are the electromagnetic interference power and Gaussian white noise power during the transmission process. j Data transmission delay and transmission energy consumption Respectively expressed as

[0158]

[0159] In the formula, A j (t) = Z j (t)y j (t) is d j The amount of data transmitted to the gateway, that is, the amount of data arriving at the edge queue.

[0160] 3) Edge data processing model

[0161] The gateway maintains a data queue for each terminal to store undecompressed data. j For example, the input of the queue in the tth time slot is the amount of data A transmitted to the gateway. j (t), the output of the tth time slot is the amount of data U processed by edge decompression j (t). Therefore, d j The data queue evolution formula is:

[0162] Q j (t+1)=Q j (t)-U j (t)+A j (t) (8)

[0163]

[0164] In the formula, The number of CPU cycles required to decompress each bit of data. Decompress and process for the gateway j CPU frequency for queue data.

[0165] Based on Little's law, the edge side d j Data queuing delay of the queue Expressed as

[0166]

[0167] In the formula, Q j The average data arrival rate (t) is expressed as

[0168]

[0169] 4) End-to-end delay and network operation cost model

[0170] In the synaesthesia synergy network, d j The end-to-end delay is composed of the data compression delay Data transmission delay Data queuing delay Among them, data compression and data transmission are parallel processes, and should be taken into account in the end-to-end delay composition. and The larger value between . Expressed as

[0171]

[0172] The cost of running a collaborative network Data compression energy consumption and data transmission energy consumption Related to, expressed as

[0173]

[0174] In the formula, ξ j is the cost coefficient.

[0175] (2) Optimization Problem Construction

[0176] In order to maintain the operating level of the distribution business carried by each acquisition terminal, the terminal must complete a certain number of data acquisition, compression and transmission within the specified time period. The total optimization time is further divided into I time periods, each of which consists of T 0 It is composed of consecutive time slots, that is, T = IT 0 The relationship between the i-th time period and the time slot is The terminal scheduling frequency constraint is

[0177]

[0178] Where N j is d j The lower limit of the number of scheduled times.

[0179] Aiming at the low-latency interactive demand for high-frequency data collection in the distribution network, the weighted sum of the end-to-end delay of the distribution business data and the operation cost of the synergistic collaborative network is minimized through synergistic collaborative optimization, that is, joint optimization of terminal scheduling, data compression ratio, bandwidth allocation and power control. The optimization problem is expressed as

[0180]

[0181] In the formula, α is the weight of the network operation cost, which is used to balance the order of magnitude and achieve the trade-off of optimization objectives. max is the total available bandwidth for data transmission, P min and P max are the upper and lower limits of the transmission power respectively. j,max and j,min They are d j The upper and lower limits of the compression ratio. 1 and C 2 is the terminal scheduling indicator variable constraint, C 3 is the transmission power constraint, C 4 is the bandwidth allocation constraint, C5 Select a constraint for the compression ratio. is the long-term terminal scheduling frequency virtual queue, and V represents the weight of the virtual queue, which is used to achieve a trade-off between minimizing queue drift and minimizing the weighted sum of end-to-end delay and network operation cost.

[0182] 2. Synaesthesia-cooperative intelligent resource allocation algorithm for high-frequency acquisition and scheduling frequency perception

[0183] This paper proposes a synaesthesia collaborative intelligent resource allocation algorithm for high-frequency acquisition scheduling frequency perception. First, a Markov model is constructed; second, the DQN loss function is improved for distribution network service acquisition needs; then, a DQN network learning and training method is proposed; finally, a transmission power and bandwidth distributed optimization method based on the ADMM algorithm is proposed.

[0184] (1) Constructing a Markov model

[0185] P1 is modeled as a Markov Decision Process (MDP) model, and the agent maintained by the gateway is composed of an evaluation network Target Network And experience replay pool and other parts. The MDP model consists of three elements: state space, action space and reward, which are described in detail as follows.

[0186] State space: The state space is defined as the set of terminal states, expressed as Terminal d j Status S j (t) includes the amount of perceived data, available CPU frequency, compression complexity, edge queue backlog, terminal scheduling frequency virtual queue deficit, and is expressed as

[0187]

[0188] Action space: Define the action space as the terminal scheduling decision x j (t) and data compression ratio decision y j (t) composed of a set Expressed as

[0189] M j (t) = {x j (t),y j (t)} (17)

[0190] Reward: Define reward θ(t) as the opposite of the optimization objective, expressed as

[0191]

[0192] (2) Improvement of DQN loss function

[0193] The DQN loss function reflects the learning performance of the agent. On the basis of traditional DQN, the algorithm's perception of scheduling frequency constraints is improved by introducing a penalty term that measures the satisfaction of long-term terminal scheduling frequency constraints in the loss function calculation. Specifically, the worse the constraint satisfaction, the larger the penalty term, indicating that the actual benefit obtained under the current action is lower than the estimate. The proposed algorithm will constrain such actions and encourage the DQN network to explore or develop other better actions, achieving more reasonable synaesthesia collaborative terminal scheduling and data compression decisions.

[0194] The loss function including the penalty term is calculated as

[0195]

[0196] In the formula, 1{t=iT 0} is an indicator function, if and only if t = iT 0 The value is 1 when , otherwise it is 0. γ is the discount factor. β(t) is the experience dataset sampled from the experience replay pool. Is the penalty factor. Based on the reward, each T 0 If the terminal scheduling frequency constraint is not met in a time slot, a penalty item needs to be subtracted from the original reward.

[0197] (3) DQN network learning

[0198] Step 1: At the beginning of each time slot, the state space is input into the evaluation network In the example above, we can obtain the Q value of taking different actions. And select the action with the largest Q value, which is

[0199]

[0200] The agent performs action M * (t).

[0201] Step 2: At the end of each time slot, the agent observes the end-to-end delay, network operation cost, and virtual queue deficit performance, calculates the reward θ(t) according to equation (18), and updates Q j (t) and And transfer the state to S(t+1). Then, the empirical data [S(t),M * (t),θ(t),S(t+1)] are stored in the experience replay pool.

[0202] Step 3: The agent calculates the loss function based on formula (19) and updates the evaluation network according to the gradient descent method. Every T 0 The time slot will target the network and Alignment.

[0203] (4) Distributed optimization of transmission power and bandwidth based on ADMM algorithm

[0204] Based on the known terminal scheduling strategy and data compression ratio, the optimization problem P1 is further transformed into

[0205]

[0206] sC 4 ~C 5 (twenty one)

[0207] Step 1: To simplify the original constraints, decompose the problem into a form that is easy to solve by Lagrange multipliers and introduce auxiliary variables w j =B j (t) and The optimization problem is transformed into

[0208]

[0209] Step 2: Establish the augmented Lagrangian function of P3, expressed as

[0210]

[0211] In the formula, λ j , μ j is the Lagrange multiplier, ρ is the penalty factor, is the L2 norm. are the vectors corresponding to each variable set.

[0212] Select Initial Value

[0213] Step 3: Update auxiliary variables, each terminal transmission power, bandwidth and Lagrange multiplier, expressed as

[0214]

[0215] P min ≤g j ≤P max (26)

[0216]

[0217] Where n is the number of iterations.

[0218] Step 4: Increase the number of iterations by 1.

[0219] Step 5: Determine the convergence of the algorithm. The algorithm terminates if the following conditions are met:

[0220]

[0221] In the formula, ε 1 , ε 2 is the preset tolerance.

[0222] Finally, the gateway will converge to the transmission power obtained in the last iteration. and bandwidth as the allocation decision for the current time slot.

[0223] It can be understood that the present invention proposes a method for constructing a high-frequency collection and low-latency interactive network model of synergistic distribution network data, jointly optimizes terminal scheduling, data compression ratio, bandwidth allocation and power control, and constructs an optimization problem with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the operation cost of the synergistic network, so as to achieve a balance between delay performance and network operation cost. In addition, the Lyapunov theory is used to decouple long-term optimization problems, and a preliminary solution to the problem of high-complexity multi-dimensional resource collaborative allocation is achieved.

[0224] It can be understood that the present invention proposes a synaesthesia collaborative intelligent resource allocation algorithm for high-frequency acquisition scheduling frequency perception, which uses a segmented processing method to gradually solve the high-complexity multi-dimensional resource collaborative allocation problem. At the same time, by introducing the penalty term that reflects the satisfaction of the long-term terminal scheduling frequency constraint into the calculation of the loss function, the scheduling frequency perception ability in the strategy learning process is improved, avoiding the situation of violating the problem constraints in exchange for maximizing benefits.

[0225] Embodiment 3:

[0226] This embodiment provides a terminal device, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the high-frequency collection and low-latency interactive method of distribution network data based on synaesthesia collaboration, including the following steps:

[0227] Construct a synergetic collaborative network model, with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the operation cost of the synergetic collaborative network, construct an optimization problem, apply the Lyapunov optimization method, transform the constructed optimization problem into a form suitable for solving, and jointly optimize dynamic scheduling, resource allocation and power control;

[0228] Based on the optimization problem, a Markov model is constructed, and the DQN loss function is improved to adapt to the distribution network service collection needs. The DQN network learning and training method is designed to optimize terminal scheduling and data compression decisions. Finally, through the transmission power and bandwidth distributed optimization method based on the ADMM algorithm, the network resource allocation is further optimized to achieve the optimal data collection strategy with low latency, low cost and balanced scheduling.

[0229] Embodiment 4:

[0230] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0231] The processor may load and execute one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the method for high-frequency collection of distribution network data and low-latency interaction based on synaesthesia collaboration in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by the processor as follows:

[0232] Construct a synergetic collaborative network model, with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the operation cost of the synergetic collaborative network, construct an optimization problem, apply the Lyapunov optimization method, transform the constructed optimization problem into a form suitable for solving, and jointly optimize dynamic scheduling, resource allocation and power control;

[0233] Based on the optimization problem, a Markov model is constructed, and the DQN loss function is improved to adapt to the distribution network service collection needs. The DQN network learning and training method is designed to optimize terminal scheduling and data compression decisions. Finally, through the transmission power and bandwidth distributed optimization method based on the ADMM algorithm, the network resource allocation is further optimized to achieve the optimal data collection strategy with low latency, low cost and balanced scheduling.

[0234] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0235] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0236] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0237] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0238] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0239] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A method for high-frequency collection and low-latency interaction of distribution network data, characterized in that: The method comprises: Construct a synergetic collaborative network model, with the goal of minimizing the weighted sum of the end-to-end delay of distribution business data and the operation cost of the synergetic collaborative network, construct an optimization problem, apply the Lyapunov optimization method, transform the constructed optimization problem into a form suitable for solving, and jointly optimize dynamic scheduling, resource allocation and power control; Based on the optimization problem, a Markov model is constructed, and the DQN loss function is improved to adapt to the distribution network service collection needs. The DQN network learning and training method is designed to optimize terminal scheduling and data compression decisions. Finally, through the transmission power and bandwidth distributed optimization method based on the ADMM algorithm, the network resource allocation is further optimized to achieve the optimal data collection strategy with low latency, low cost and balanced scheduling.

2. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 1, characterized in that: The synaesthesia collaborative network model includes: a collection terminal scheduling model, a perception data compression and transmission model, an edge data processing model, and an end-to-end delay and network operation cost model; The total optimization time is divided into T time slots, and the set is expressed as The length of each time slot is τ0, and each time slot includes three processes: perception data collection, data compression and transmission, and edge data processing; Among them, building a collection terminal scheduling model includes: Determine the number of terminals at the terminal layer as J, assuming that the gateway schedules I in each time slot th <J terminals collect and upload data, and determine the terminal scheduling indicator variable as x j (t), x j (t) = 1 indicates terminal d j Scheduled in the tth time slot, x j (t) = 0 means not scheduled, and the terminal d of the tth time slot is determined. j The amount of sensory data collected is a j (t), the amount of waiting compression is expressed as: Z j (t)=x j (t)a j (t) (1) Among them, the construction of perception data compression and transmission model includes: Considering the redundancy of the original data, the acquisition terminal needs to compress the data to determine d j The data compression rate is It is related to the data compression ratio y j (t) is related to, expressed as: In the formula, is the compression complexity, that is, the number of CPU cycles required to compress each bit of data, is d j Available CPU frequency, d j Data compression delay And data compression energy consumption Respectively expressed as: In the formula, ζ j is d j Energy consumption coefficient; During the data compression process, the data is transmitted to the gateway synchronously and in parallel. j Data transmission rate R between the gateway j (t) is expressed as: In the formula, B j (t) is d j The transmission bandwidth, h j (t) is the channel gain, e j (t) and ζ0 are the electromagnetic interference power and Gaussian white noise power during the transmission process, respectively. j Data transmission delay and transmission energy consumption Respectively expressed as: In the formula, A j (t) = Z j (t)y j (t) is d j The amount of data transmitted to the gateway, that is, the amount of data arriving at the edge queue; Among them, building an edge data processing model includes: The gateway maintains a data queue for each terminal to store undecompressed data. j For example, the input of the queue in the tth time slot is the amount of data A transmitted to the gateway. j (t), the output of the tth time slot is the amount of data U processed by edge decompression j (t), therefore, d j The data queue evolution formula is: Q j (t+1)=Q j (t)-U j (t)+A j (t) (8) In the formula, The number of CPU cycles required to decompress each bit of data, Decompress and process for the gateway j CPU frequency of queue data; Based on Little's law, the edge side d j Data queuing delay of the queue It is expressed as: In the formula, Q j The average data arrival rate (t) is expressed as: The end-to-end delay and network operation cost model is constructed, including: In the synaesthesia synergy network, d j The end-to-end delay is composed of the data compression delay Data transmission delay Data queuing delay In which data compression and data transmission are parallel processes, and the end-to-end delay should be taken into account. and The larger value between It is expressed as: The cost of running a collaborative network Data compression energy consumption and data transmission energy consumption Related, expressed as: In the formula, ξ j is the cost coefficient.

3. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 2, characterized in that: The construction optimization problem specifically includes: In order to maintain the operating level of the distribution business carried by each acquisition terminal, the terminal must complete a certain number of data acquisition, compression and transmission within the specified time period; further divide the total optimization time into I time periods, each time period consists of T0 continuous time slots, that is, T = IT0, and the relationship between the i-th time period and the time slot is The terminal scheduling frequency constraint is: Where N j is d j The lower limit of the number of dispatches; Aiming at the low-latency interactive demand for high-frequency data collection in the distribution network, the weighted sum of the end-to-end delay of the distribution business data and the operation cost of the synergistic collaborative network is minimized through synergistic collaborative optimization, that is, joint optimization of terminal scheduling, data compression ratio, bandwidth allocation and power control. The optimization problem is expressed as: In the formula, α is the weight of the network operation cost, which is used to balance the order of magnitude and achieve the optimization target trade-off, B max is the total available bandwidth for data transmission, P min and P max are the upper and lower limits of the transmission power, y j,max and j,min They are d j The upper and lower limits of the compression ratio, C1 and C2 are the terminal scheduling indicator variable constraints, C3 is the transmission power constraint, C4 is the bandwidth allocation constraint, and C5 is the compression ratio selection constraint. is the long-term terminal scheduling frequency virtual queue, and V represents the weight of the virtual queue, which is used to achieve a trade-off between minimizing queue drift and minimizing the weighted sum of end-to-end delay and network operation cost.

4. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 1, characterized in that: The construction of the Markov model specifically includes: P1 is modeled as a Markov decision MDP model, and the agent maintained by the gateway is composed of an evaluation network Target Network As well as the experience replay pool, the MDP model includes three elements: state space, action space and reward, as follows: State space: The state space is defined as the set of terminal states, expressed as Terminal d j Status S j (t) includes the amount of perceived data, available CPU frequency, compression complexity, edge queue backlog, terminal scheduling frequency virtual queue deficit, and is expressed as: Action space: Define the action space as the terminal scheduling decision x j (t) and data compression ratio decision y j (t) composed of a set It is expressed as: M j (t)={x j (t),y j (t)} (17) Reward: Define reward θ(t) as the opposite of the optimization objective, expressed as:

5. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 1, characterized in that: The improved DQN loss function specifically includes: The DQN loss function reflects the learning performance of the intelligent agent. By introducing a penalty term to measure the satisfaction of the long-term terminal scheduling frequency constraint in the loss function calculation, the algorithm's perception of the scheduling frequency constraint is improved. Specifically, when the constraint satisfaction is worse, the penalty term is larger, indicating that the actual benefit obtained under the current action is lower than the estimate. The proposed algorithm will constrain such actions and encourage the DQN network to explore or develop other better actions, so as to achieve more reasonable synaesthesia collaborative terminal scheduling and data compression decisions. The loss function including the penalty term is calculated as: Where 1{t=iT0} is the indicator function, which takes the value of 1 if and only if t=iT0, otherwise it takes the value of 0. γ is the discount factor, β(t) is the experience data set sampled from the experience replay pool, It is a penalty factor. Based on the reward, if the terminal scheduling frequency constraint is not met in each T0 time slot, a penalty item needs to be subtracted from the original reward.

6. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 1, characterized in that: The DQN network learning specifically includes: First, at the beginning of each time slot, the state space is input into the evaluation network In the example above, we can obtain the Q value of taking different actions. And select the action with the largest Q value, which is: The agent performs action M * (t); Then, at the end of each time slot, the agent observes the end-to-end delay, network operation cost, and virtual queue deficit performance, calculates the reward θ(t) according to equation (18), and updates Q j (t) and And transfer the state to S(t+1), then, the experience data [S(t),M * (t),θ(t),S(t+1)] are stored in the experience replay pool; Finally, the agent calculates the loss function based on formula (19) and updates the evaluation network according to the gradient descent method. Every T0 time slot, the target network and Alignment.

7. A method for high-frequency collection and low-latency interaction of distribution network data according to claim 1, characterized in that: The transmission power and bandwidth distributed optimization based on the ADMM algorithm specifically includes: Based on the known terminal scheduling strategy and data compression ratio, the optimization problem P1 is further transformed into: Step 1: To simplify the original constraints, decompose the problem into a form that is easy to solve by Lagrange multipliers and introduce auxiliary variables w j =B j (t) and The optimization problem is transformed into: Step 2: Establish the augmented Lagrangian function of P3, expressed as: In the formula, λ j , μ j is the Lagrange multiplier, ρ is the penalty factor, is the L2 norm, are the vectors corresponding to each variable set; Select Initial Value Step 3: Update the auxiliary variables, the transmission power of each terminal, the bandwidth and the Lagrange multiplier, expressed as: Where n is the number of iterations; Step 4: The number of iterations increases by 1; Step 5: Determine the convergence of the algorithm. The algorithm terminates if the following conditions are met: In the formula, ε1 and ε2 are the preset tolerances; Finally, the gateway will converge to the transmission power obtained in the last iteration. and bandwidth as the allocation decision for the current time slot.

8. A distribution network data high-frequency collection and low-latency interactive system, characterized in that: The system comprises: a device layer, a perception layer, an edge layer and a master station layer; The equipment layer: including distributed photovoltaics, distributed wind turbines, smart charging piles, energy storage batteries and various electrical equipment such as power distribution equipment; The perception layer includes a plurality of high-frequency distribution network data collection terminal devices, which are used to perceive the key data of electrical quantity, state quantity and network environment quantity during the operation of various electrical equipment, and compress and package the perceived data according to the received scheduling and compression scheme, and upload the compressed data to the edge layer according to the received bandwidth and transmission power allocation strategy; The edge layer: includes a high-frequency data collection and low-latency interactive gateway device for distribution network based on synaesthesia collaboration and a 5G base station; the 5G base station provides communication coverage for the perception layer, and the high-frequency data collection and low-latency interactive gateway device for distribution network based on synaesthesia collaboration is responsible for decompressing and processing the uploaded compressed data, and uploading it to the main station layer through optical fiber, providing low-latency data support for the operation of distribution business, and issuing strategies for the perception layer; The master station layer includes a power grid control center, which is responsible for supporting cross-regional coordinated control of distribution networks, panoramic monitoring of distribution networks, and main and distribution coordinated control business operations.

9. A distribution network data high frequency collection and low latency interactive system according to claim 8, characterized in that: The distribution network data high-frequency collection terminal device of the perception layer in the synaesthesia-coordinated distribution network data high-frequency collection and low-latency interactive system comprises: a collection module, a compression module, a scheduling perception module and a communication module; The acquisition module includes a high-frequency data acquisition terminal for distribution network deployed on various electrical equipment, which is responsible for sensing the key data of electrical quantity, state quantity and network environment quantity during the operation of various electrical equipment; The compression module is responsible for compressing and packaging the original data according to the instruction of the scheduling perception module; The scheduling perception module is responsible for determining whether to upload data according to the decision issued by the edge layer; The communication module is responsible for transmitting data through the 5G channel, receiving the scheduling and compression schemes and bandwidth and transmission power allocation decisions fed back by the edge layer, and sending compressed and packaged key equipment operation data to the edge layer.

10. A distribution network data high frequency collection and low latency interactive system according to claim 8, characterized in that: The edge layer's synaesthesia-based distribution network data high-frequency collection and low-latency interactive gateway device includes: a synaesthesia-based model building module, an optimization problem building and transformation module, a Q network learning module, a scheduling frequency-aware loss function improvement module, an action selection module, an augmented Lagrangian function building module, a distributed optimization iteration module, and a communication module; The synaesthesia collaboration model building module is responsible for building the acquisition terminal scheduling model, the perception data compression and transmission model, the edge data processing model, and the end-to-end delay and network operation cost model; The optimization problem construction and transformation module is responsible for constructing the weighted sum optimization problem of minimizing the end-to-end delay of distribution business data and the operation cost of the synergy collaborative network based on the synergy collaborative model, and transforming the problem based on Lyapunov; The Q network learning module is responsible for maintaining a set of DQN networks and experience replay pools for each terminal; The scheduling frequency-aware loss function improvement module is responsible for introducing a penalty term reflecting the terminal acquisition frequency to improve the calculation process of the loss function; The action selection module is responsible for selecting the action with the largest Q value based on the DQN network as the terminal scheduling and data compression ratio selection decision; The augmented Lagrangian function construction module is responsible for introducing auxiliary variables and constructing an augmented Lagrangian function for the joint optimization problem of transmission power and bandwidth; The distributed optimization iteration module is responsible for updating auxiliary variables, transmission power of each terminal, bandwidth and Lagrange multipliers until the algorithm converges; The communication module is responsible for communicating with the 5G base station, issuing terminal scheduling, data compression ratio scheme, and bandwidth and transmission power allocation decisions, receiving data uploaded by the high-frequency acquisition terminal, and uploading data to the main station.

Images