6G Power Semantic Communication Method, Device and System Based on Sensing and Communication Integration

Abstract

The present invention discloses a 6G electric power semantic communication method, device and system based on synaesthesia integration, including: constructing a new information timeliness measurement index; considering the constraints of distribution network hierarchical collaborative regulation on information timeliness, constructing a joint optimization problem, the goal is to minimize the peak semantic age of the distribution network; according to the differences of the optimization subjects, the original problem is divided into two sub-problems; constructing the terminal scheduling problem as a Markov decision process, through the Top‑N <supgt;2< / supgt;实现优化;将数据采集频率与语义压缩比联合优化问题建模为Multi‑Armed Bandit Problem,构建六层模糊网络来挖掘摇臂与运行状态的适配度;基于归一化后的适配度计算摇臂效用,终端选择效用值最小的采集频率和语义压缩比组合,终端基于边缘服务器反馈的峰值语义年龄信息更新统计学习参数。提高了电力系统的运行效率和稳定性。

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and particularly relates to a 6G power semantic communication method, device, and system based on integrated communication and sensing. Background Art

[0002] With the large-scale development of distributed power sources, microgrids, and virtual power plants, the distribution network is evolving from a vertical control mode to a hierarchical collaborative control mode of province-region-distribution-microgrid and regional autonomy, posing higher requirements for the data perception and real-time reliable transmission capabilities of 6G wireless communication networks. 6G power semantic communication uses semantics as the basic unit of information representation, realizing a paradigm change from symbol transmission to semantic transmission, and providing a solution to the challenge of low-latency transmission of massive data in 6G communication. 6G power semantic communication can be further combined with the hierarchical collaborative control service of the distribution network, further compressing the power semantic information based on service characteristics, and significantly reducing the data transmission volume while ensuring the smooth implementation of the service.

[0003] The core of 6G power semantic communication is integrated communication and sensing resource allocation. By scheduling Internet of Things (IoT) terminals to collect and monitor the operating environment and status of electrical equipment, semantic extraction and semantic compression are performed on the collected raw data based on power service characteristics, and the compressed data is uploaded to the edge side. The edge side decodes and processes the semantic information received by the 6G base station through an edge server, effectively reducing the communication resources occupied during data transmission and reducing the transmission latency of massive data in 6G communication. For the problem that it is difficult to construct a mechanism model for integrated communication and sensing, reinforcement learning mines historical observation data to find the complex mapping relationship between resource allocation decisions and optimization objectives.

[0004] However, how to achieve the integrated allocation of sensing and communication resources such as data acquisition frequency, terminal scheduling, and semantic compression ratio, and ensure the information timeliness requirements of the hierarchical collaborative control service of the distribution network remains a core technical issue, mainly facing the following challenges. First, traditional power semantic communication lacks a method for measuring information timeliness. Delay only measures the system performance narrowly from the transmission perspective and cannot effectively characterize the entire life cycle of semantic communication covering information acquisition, compression, transmission, and decoding. Moreover, timeliness indicators such as age of information ignore the underlying meaning of the power service information source and only measure from the perspective of successful information reception, making it difficult to reflect the semantic understanding performance of the receiving end. Second, the coupling of sensing and communication makes resource allocation an NP-hard problem. Although increasing the acquisition frequency can accurately capture the key features of important services, the large amount of data generated by high-frequency acquisition will exceed the semantic compression processing capacity of the terminal side and the channel capacity, resulting in a sharp increase in queue backlog and deterioration of the end-to-end delay performance. Finally, existing reinforcement learning methods are difficult to achieve complex learning problems in resource allocation in the hierarchical collaborative control system of the distribution network, with poor generalization and interpretability, limiting the practical application of the system. Summary of the Invention

[0005] To solve the technical problems existing in the background art, the present invention aims to provide a 6G power semantic communication method, device, and system based on integrated sensing and communication, which reduces data transmission delay through 6G semantic communication, realizes the reasonable allocation of integrated resources for sensing and semantic communication, and ensures the information timeliness of the hierarchical collaborative control service of the distribution network.

[0006] To solve the technical problems, the technical solution of the present invention is as follows:

[0007] A 6G power semantic communication method based on integrated sensing and communication, the method includes:

[0008] Based on the constructed sensing model and communication model on the terminal side, construct a new information timeliness metric suitable for power semantic communication, namely peak semantic age;

[0009] Considering the constraints of the hierarchical collaborative control of the distribution network on information timeliness, construct an optimization problem that jointly optimizes sensing frequency selection, terminal scheduling, and semantic compression ratio to minimize the peak semantic age of the distribution network;

[0010] Based on the differences of the optimization entities, split the original problem into a side terminal scheduling sub-problem and a joint optimization sub-problem of acquisition frequency selection and semantic compression ratio on the terminal side;

[0011] Construct the terminal scheduling problem as a Markov decision process and optimize the terminal scheduling through Top-N 2 Realize the optimization of terminal scheduling;

[0012] The problem of jointly optimizing the data acquisition frequency and semantic compression ratio is modeled as a Multi-Armed Bandit (MAB) problem, and a six-layer fuzzy network is constructed to mine the fitness between the arms and the operating states.

[0013] Based on the normalized fitness, the utility of the arms is calculated, and the terminal selects the combination of the acquisition frequency and semantic compression ratio with the minimum utility value. The terminal updates the statistical learning parameters based on the peak semantic age information fed back by the edge server.

[0014] Furthermore, the construction of the perception model on the terminal side specifically includes:

[0015] Suppose there are I Internet of Things terminals, and the terminal set is represented as Determine the terminal d i The acquisition frequency at the t-th time slot is f i (t), determine f i,max and f i,min are the upper and lower limits of the acquisition frequency respectively. The value range of the acquisition frequency is discretized into M levels, denoted as where the acquisition frequency of the m-th level is denoted as

[0016] Furthermore, the construction of the communication model specifically includes:

[0017] Determine the terminal scheduling variable as x i (t), x i (t) = 1 indicates that at the t-th time slot, the terminal d i is scheduled to transmit the compressed semantics to the edge server; suppose that at the t-th time slot, the terminal d i is scheduled to upload the compressed semantic information; determine the terminal d i The semantic compression ratio at the t-th time slot is the ratio of the amount of compressed data to the amount of data before compression, denoted as y i (t); determine y i,max and y i,min are the upper and lower limits of the semantic compression ratio respectively. The value range of the semantic compression ratio is discretized into N levels, denoted as where the semantic compression ratio of the n-th level is denoted as

[0018] Considering that the compressed semantics are quantized into bits and transmitted through 6G, at the t-th time slot, the transmission rate from the terminal d i to the edge server is:

[0019]

[0020] In the formula, B i is the transmission bandwidth between the terminal d i and the edge server, Pi is d i 's transmission power, h i (t) is d i 's channel gain with the edge server, N 0 is the noise power spectral density, N i (t) is d i 's electromagnetic interference power;

[0021] The terminal d i caches the collected data locally to form a data queue, and the data queue backlog is:

[0022] Q i (t + 1) = Q i (t) + f i (t)a i + L i (t) - U i (t) (2)

[0023] In the formula, the input of the queue is Q i (t), f i (t)a i and L i (t); f i (t) represents the collection frequency of the terminal d i at the t-th time slot, a i is the amount of data collected per single collection, L i (t) is the amount of data that needs to be retransmitted due to semantic decoding errors, U i (t) is the amount of data transmitted, representing the output of the queue, and the calculation formula is:

[0024]

[0025] In the formula, [·] - represents the minimum function, represents the semantic compression rate of the terminal d i at the t-th time slot, and the calculation formula is:

[0026]

[0027] In the formula, is the computing resource required for the terminal d i to compress one unit of bit semantic data, is the computing resource used by the terminal d i to perform the semantic compression task.

[0028] Furthermore, the construction of the new information timeliness metric specifically includes:

[0029] Construct a mapping function Indicate that the edge server processes terminal d at the t-th time slot i The e-th semantic data packet of is transmitted at the t'-th time slot, determine y i (t) is for terminal d i At the semantic compression ratio of terminal d at the t-th time slot, the decoding delay of the e-th semantic data packet is:

[0030]

[0031] In the formula, Indicates the computing resources required for the edge server to decode unit-bit semantic data in the t time slot Indicates the computing resources used by the edge server to perform the semantic decoding task;

[0032] Determine S i (t,e) is the decoding success rate of the e-th semantic data packet at the t-th time slot. The decoding success rate of the semantic data packet can be modeled as the weighted sum of two exponential functions, expressed as:

[0033]

[0034] In the formula, β 1 、β 2 、β 3 、β 4 Indicates model parameters;

[0035] Determine that the semantic successful decoding variable is Z i (t,e), Z i (t,e) = 1 indicates that the e-th semantic data packet of terminal d i Is successfully decoded at the t-th time slot, otherwise Z i (t,e) = 0; Determine PAoS i (t,e) represents the peak semantic age of the e-th semantic data packet of terminal d i At the t-th time slot, the calculation formula is:

[0036]

[0037] In the formula, ω i (t,e) represents the importance of the semantic data packet, reflecting the importance degree of the service data characteristics corresponding to the semantics for the hierarchical collaborative control of the distribution network; in the brackets, the first item δ i (t,e) represents the end-to-end delay, the second item Represents the waiting delay caused by semantic decoding failure, O e Refers to the index of the next successfully decoded packet after the e-th packet, the third item Δ i (t,O e ) represents the semantic decoding delay;

[0038] Determine the terminal d i The peak semantic age at the t-th time slot is the maximum value among the peak semantic ages of all data packets, denoted as:

[0039] PAoS i (t) = max e {PAoS i (t, e)} (8)

[0040] Furthermore, by jointly optimizing the sensing frequency selection, terminal scheduling, and semantic compression ratio, minimize the peak semantic age of the distribution network; determine the optimization variables as f = {f i (t)}, x = {x i (t)}, y = {y i (t)}, and the optimization problem is constructed as:

[0041]

[0042] In the formula, C 1 is the data acquisition frequency constraint; C 2 is the terminal scheduling constraint, indicating that an edge server can receive data from at most q terminals simultaneously; C 3 is the semantic compression ratio constraint;

[0043] Based on the differences of the optimization entities, the original problem P1 is split into the edge-side terminal scheduling sub-problem SP1 and the end-side acquisition frequency selection and semantic compression ratio joint optimization sub-problem SP2.

[0044] Furthermore, construct the terminal scheduling problem as a Markov decision process, and optimize the terminal scheduling through Top-N 2 The specific implementation includes:

[0045] At the t-th time slot, the state space of the edge server is defined as the set of the state spaces of each terminal, denoted as where the state space of the terminal d i is G i (t) is the data queue backlog of the edge-side terminal d i at the t-th time slot;

[0046] At the t-th time slot, the action space of the edge server is defined as the set of terminal scheduling decisions, denoted as

[0047]

[0048] The optimization problem SP1 is a minimization problem. Define the cost function at the t-th time slot as the optimization objective of SP1, that is

[0049] Using the DQN network to mine and fit the cost of the scheduling terminal d i That is, the Q value, and minimize the cost function by scheduling the q terminals with the smallest Q value. The implementation steps of the Top-N 2 algorithm include:

[0050] The edge server inputs the terminal numbers and their state spaces into the DQN network to obtain the Q value

[0051]

[0052] Arrange the terminals in ascending order based on the Q value and select the q terminals with the smallest Q value for scheduling, which can be expressed as:

[0053]

[0054] At the end of the time slot, the edge server updates the queues Q i (t + 1), L i (t + 1), G i (t + 1), observe the PAoS and semantic importance of each terminal, and calculate the loss function, which is expressed as:

[0055]

[0056] The edge server updates the DQN network based on the loss function and the gradient descent method

[0057] Furthermore, considering the limited computing power of the terminal, the joint optimization problem of data collection frequency and semantic compression ratio is modeled as an MAB problem, and a lightweight UCB algorithm is used to solve it; each arm of the slot machine corresponds to a combination scheme of data collection frequency and semantic compression ratio; the UCB algorithm balances exploration and exploitation. On the one hand, it uses statistical information to select the arm with the best known empirical performance, and on the other hand, it explores new arms to obtain better rewards, so as to obtain the maximum reward; it includes three steps: hybrid-driven arm utility evaluation, action formulation, and statistical learning parameter update;

[0058] The specific process includes:

[0059] Step 1: Hybrid-driven arm utility evaluation:

[0060] The utility of the arm includes two parts. One is the fitness between the arm constructed based on knowledge-driven fuzzy learning and the operating states of electrical equipment, i.e., the primary side, and the terminal, i.e., the secondary side; the second is the upper confidence bound of the arm performance constructed based on statistical-driven reinforcement learning;

[0061] 1) Fitness calculation

[0062] Construct a six-layer fuzzy network to mine the fitness between the rocker arm and the operating state;

[0063] The first layer is the state input layer, which consists of six neurons, corresponding to six elements in the terminal-side state space respectively, namely the average semantic importance of the data collected by the terminal d at time slot t-1 i Queue backlog Q (t), the data to be retransmitted in this time slot L i (t), the terminal scheduling result x i (t), the acquisition frequency action f i (t), the semantic compression ratio action y i (t); i (t);

[0064] The second layer is the membership degree calculation layer, which is used to calculate the membership degree of each state for different fuzzy levels. The membership degree γ i,j,k (t) reflects the degree to which the j-th state of the terminal d at time slot t belongs to the k-th fuzzy level. The calculation formula is: i (t) reflects the degree to which the j-th state of the terminal d at time slot t belongs to the k-th fuzzy level. The calculation formula is:

[0065]

[0066] In the formula, μ i,j,k and φ i,j,k are the central value and the diffusion parameter of the k-th fuzzy level for state j respectively. The membership degree calculation layer converts the specific state value into a fuzzy value;

[0067] The third layer is the firing strength calculation layer. The firing strength of the fuzzy rule is expressed as the product of the membership degrees of each state, which is expressed as:

[0068]

[0069] The fourth layer is the normalization layer, which is used to normalize the firing strength. The normalized firing strength is expressed as

[0070] The fifth layer is the fully connected layer, which is used to calculate the activation strength of each fuzzy rule, which is expressed as:

[0071]

[0072] In the formula, is the activation strength of the fuzzy rule selecting the m-th acquisition frequency action and the n-th semantic compression action at time slot t; v i,j (t) represents the weight of the neural network; b i,j (t) is the bias;

[0073] The sixth layer is the output layer, which is used to output the adaptation degree result between the electrical device and the terminal, expressed as:

[0074]

[0075] In the formula, W is the number of fuzzy rules, which is equal to the product of the number of fuzzy levels of each state;

[0076] 2) Utility calculation

[0077] To balance the influence of knowledge-driven and statistics-driven, the output adaptation degree result is normalized, expressed as:

[0078]

[0079] In the formula, the parameter ξ 1 →ξ 0 +,ξ 0 →0+ normalizes the adaptation degree to between (0, 1); based on the normalized adaptation degree The utility value is calculated as:

[0080]

[0081] In the formula, θ i,m,n (t) represents the empirical performance of the terminal d i selecting the m-th acquisition frequency and the n-th semantic compression ratio up to the t-th time slot; α i,m,n (t) represents the number of times the m-th acquisition frequency and the n-th semantic compression ratio have been selected up to the t-th time slot;

[0082] Step 2: Action formulation:

[0083] Determine the action formulation variable as λ i,m,n (t); λ i,m,n (t) = 1 indicates that the terminal d i selects the m-th acquisition frequency and the n-th semantic compression ratio at the t-th time slot, otherwise λ i,m,n (t) = 0; The terminal selects the combination of the acquisition frequency and the semantic compression ratio with the minimum utility value (m * , n * ), expressed as:

[0084]

[0085] Step 3: Update of statistical learning parameters:

[0086] At the end of the time slot, the terminal updates the statistical learning parameters based on the PAoS information fed back by the edge server as:

[0087]

[0088] A 6G power semantic communication device based on integrated sensing and communication, the device comprising: a power semantic communication IoT terminal, an electrical device, a 6G base station, and a power semantic communication computing gateway;

[0089] The electrical device is electrically connected to the power semantic communication IoT terminal, the power semantic communication IoT terminal is signal-connected to the 6G base station, and the power semantic communication computing gateway is signal-connected to the 6G base station;

[0090] Among them, the power semantic communication IoT terminal collects and monitors the operating environment and status of the electrical device, extracts and compresses the semantics of the collected data based on power service characteristics, and uploads the compressed data to the power semantic communication computing gateway through the 6G base station. The power semantic communication computing gateway decodes and processes the received semantic information and provides data for power services.

[0091] A 6G power semantic communication system based on integrated sensing and communication, the system comprising: an edge layer and a terminal layer; the terminal layer includes the power semantic communication IoT terminal and the electrical device described above; the edge layer includes the 6G base station and the power semantic communication computing gateway described above;

[0092] The power semantic communication IoT terminal includes: a data acquisition module, a semantic encoding module, a semantic compression module, a channel encoding module, a semantic importance recognition module, a communication module, a queue backlog perception module, an integrated sensing and communication computing module, and a power supply module;

[0093] The data acquisition module is used to collect the operating environment and status data of the electrical device;

[0094] The semantic encoding module is used to extract semantic information from the above-mentioned originally collected data based on service characteristics;

[0095] The semantic compression module is used to compress the extracted semantic information;

[0096] The channel encoding module is used to add redundant information to improve the reliability of semantic information transmission in a complex channel environment;

[0097] The semantic importance recognition module is used to identify the semantic importance of the collected data;

[0098] The communication module is used to send the collected data and semantic importance information to the power semantic communication computing gateway, and receive the information on whether the decoding is successful and the terminal scheduling information issued by the power semantic communication computing gateway;

[0099] The queue backlog perception module is used to calculate the queue backlog of the collected data;

[0100] The cross-sensory integration computing module is used to optimize the terminal sensing frequency selection and semantic compression ratio through knowledge-statistics hybrid-driven fuzzy reinforcement learning, and send control information to the data acquisition module and the semantic compression module;

[0101] The power supply module supplies power to the data acquisition module, the semantic encoding module, the semantic compression module, the channel encoding module, the semantic importance recognition module, the communication module, the queue backlog perception module, and the cross-sensory integration computing module;

[0102] The power semantic communication computing gateway includes: a channel decoding module, a semantic decoding module, a communication module, a peak semantic age calculation module, a terminal scheduling module, a data storage module, a data processing module, and a power supply module;

[0103] The channel decoding module is used to receive signals and perform symbol detection to recover the transmitted compressed semantic information;

[0104] The semantic decoding module is used to convert the information recovered by the channel decoding module back into semantic concepts related to the task;

[0105] The communication module is used to receive the semantic information received by the 6G base station, and send the information on whether the decoding is successful and the terminal scheduling information to the power semantic communication IoT terminal;

[0106] The peak semantic age calculation module is used to calculate the peak semantic age of the received data;

[0107] The terminal scheduling module is used to utilize the DQN network to mine and fit the cost of scheduling terminals, select appropriate terminals to minimize the terminal scheduling cost, and send the terminal scheduling information to each power semantic communication IoT terminal;

[0108] The data storage module is used to store the electrical equipment operation status data information collected by the power semantic communication IoT terminal;

[0109] The data processing module is used to utilize the received electrical equipment operation status data information to conduct island operation judgment, fault handling, precise load shedding, and regional autonomous control of power services;

[0110] The power supply module supplies power to the channel decoding module, the semantic decoding module, the communication module, the peak semantic age calculation module, the terminal scheduling module, the data storage module, and the data processing module.

[0111] A computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in any one of the above.

[0112] Compared with the prior art, the advantages of the present invention are as follows:

[0113] 1. The present invention proposes a 6G communication resource allocation model integrating communication and sensing, constructs a new information timeliness metric for power semantic communication covering the entire life cycle of information acquisition, semantic compression, semantic transmission, and semantic parsing, considers the strict requirements of hierarchical collaborative control of the distribution network for information timeliness, and realizes the optimization of integrated communication and sensing resource allocation by jointly optimizing sensing frequency selection, terminal scheduling, and semantic compression ratio. In addition, considering the parsing success rate of important semantic information at the receiving end, it ensures the real-time and accurate information, guarantees the information timeliness of hierarchical collaborative control of the distribution network, and provides strong technical support for power semantic communication in the 6G communication environment.

[0114] 2. The present invention proposes a terminal scheduling optimization method based on Top-N 2 By using Top-N 2 for terminal scheduling optimization, in the edge side, the deep Q network is used to deeply explore and fit the terminal scheduling cost, and the data acquisition frequency and semantic compression ratio are dynamically adjusted according to the actual operation state of the power grid, optimizing the overall configuration of sensing and communication resources, greatly improving the power system's ability to handle emergencies and complex situations, and ensuring the rapid capture and efficient transmission of key information.

[0115] 3. The present invention proposes a fuzzy reinforcement learning method driven by a hybrid of knowledge and statistics. By deeply analyzing and learning the historical data and real-time state of power grid operation, integrating the advantages of fuzzy learning and reinforcement learning, accurately predicting the changes in power grid demand, and real-time adjusting the resource allocation strategy, realizing the joint optimization of sensing frequency selection and semantic compression ratio to cope with various operating conditions and changing situations. It provides a new solution for the efficient operation and management of the power system, with important practical application value and broad development prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] Figure 1 Flowchart of a 6G power semantic communication method based on integration of communication and sensing;

[0117] Figure 2 System diagram of a 6G power semantic communication system based on integration of communication and sensing;

[0118] Figure 3 Composition diagram of the Internet of Things terminal for power semantic communication;

[0119] Figure 4 Composition diagram of the computing gateway for power semantic communication. DETAILED DESCRIPTION OF THE INVENTION

[0120] The following describes the specific embodiments of the present invention in conjunction with the embodiments:

[0121] It should be noted that the structures, ratios, sizes, etc. shown in this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the implementation conditions of the present invention. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0122] At the same time, the terms such as "upper", "lower", "left", "right", "middle", and "one" cited in this specification are also only for the convenience of clear narration, rather than used to limit the scope of implementation of the present invention. The change or adjustment of their relative relationship, without substantial change in the technical content, should also be regarded as the scope of implementation of the present invention.

[0123] Example 1:

[0124] The present invention proposes a 6G power semantic communication method based on integrated sensing and communication. First, a power 6G semantic communication resource allocation model based on integrated sensing and communication is constructed. Secondly, a resource allocation algorithm for integrated sensing and communication based on Top-N 2 and knowledge-statistics hybrid-driven fuzzy reinforcement learning is proposed. Finally, a 6G power semantic communication device system based on integrated sensing and communication is proposed, including an end-side control layer and a side-edge control layer. The process is as Figure 1 shown, and the specific implementation scheme is introduced as follows:

[0125] 1. 6G power semantic communication resource allocation model based on integrated sensing and communication

[0126] S1: The present invention proposes a 6G power semantic communication resource allocation model based on integrated sensing and communication. First, a new information timeliness metric suitable for power semantic communication is constructed, namely Peak Age of Semantic Information (PAoS), which covers the entire life cycle of information acquisition, semantic compression, semantic transmission, and semantic parsing. Secondly, considering the strict requirements of hierarchical collaborative control of the distribution network for information timeliness, an optimization problem P1 is constructed to jointly optimize the sensing frequency selection, terminal scheduling, and semantic compression ratio to minimize the peak semantic age of the distribution network. Finally, based on the differences of the optimization subjects, the original problem P1 is split into a side-edge terminal scheduling sub-problem SP1 and an end-side acquisition frequency selection and semantic compression ratio joint optimization sub-problem SP2. The specific introduction is as follows:

[0127] S1.1: Define the time slot length as τ, and the time slot set is expressed as In each time slot, the terminal optimizes the data acquisition frequency and the semantic compression ratio. The edge server schedules the terminal to upload the compressed semantic information. By jointly optimizing the data acquisition frequency, the semantic compression ratio, and the terminal scheduling, the integrated sensing and communication resource allocation is optimized to ensure the timeliness of information for hierarchical collaborative control of the distribution network.

[0128] 1) Sensing model

[0129] Suppose there are I Internet of Things terminals in total, and the terminal set is denoted as Define terminal d i The acquisition frequency at the t-th time slot is f i (t). Define f i,max and f i,min as the upper and lower limits of the acquisition frequency respectively. The value range of the acquisition frequency is discretized into M levels, denoted as where the acquisition frequency of the m-th level is denoted as

[0130] 2) Communication model

[0131] Define the terminal scheduling variable as x i (t). x i (t) = 1 means that at the t-th time slot, terminal d i is scheduled to transmit the compressed semantics to the edge server. Suppose that at the t-th time slot, terminal d i is scheduled to upload the compressed semantic information. Define the semantic compression ratio of terminal d i at the t-th time slot as the ratio of the amount of compressed data to the amount of data before compression, denoted as y i (t). Define y i,max and y i,min as the upper and lower limits of the semantic compression ratio respectively. The value range of the semantic compression ratio is discretized into N levels, denoted as where the semantic compression ratio of the n-th level is denoted as

[0132] Considering that the compressed semantics are quantized into bits and transmitted through 6G, at the t-th time slot, the transmission rate from terminal d i to the edge server is:

[0133]

[0134] In the formula, B i is the transmission bandwidth between terminal d i and the edge server, P i is the transmission power of d i , h i (t) is the channel gain between d i and the edge server, N 0is the noise power spectral density, N i (t) is d i of the electromagnetic interference power.

[0135] The terminal d i caches the collected data locally to form a data queue. The data queue backlog is:

[0136] Q i (t + 1) = Q i (t) + f i (t)a i + L i (t) - U i (t) (2)

[0137] In the formula, the input of the queue is Q i (t), f i (t)a i and L i (t). f i (t) represents the acquisition frequency of the terminal d i at the t-th time slot, a i is the amount of data collected in a single acquisition, L i (t) is the amount of data that needs to be retransmitted due to semantic decoding errors. U i (t) is the amount of data transmitted, representing the output of the queue. The calculation formula is:

[0138]

[0139] In the formula, [·] - represents the minimum function. represents the semantic compression rate of the terminal d i at the t-th time slot. The calculation formula is:

[0140]

[0141] In the formula, is the terminal d i the computing resources required to compress one unit bit of semantic data, is the terminal d i the computing resources used to execute the semantic compression task.

[0142] 3) Peak semantic age model

[0143] Construct the mapping function indicating that the e-th semantic data packet of the terminal d i processed by the edge server at the t-th time slot was transmitted at the t'-th time slot. Define y i (t) as the terminal d iIf it is the semantic compression ratio in the t-th time slot, then the decoding delay of the e-th semantic data packet is:

[0144]

[0145] In the formula, represents the computing resources required for the edge server to decode unit-bit semantic data in the t-th time slot, represents the computing resources used by the edge server to perform the semantic decoding task.

[0146] Define S i (t, e) as the decoding success rate of the e-th semantic data packet in the t-th time slot. The decoding success rate of the semantic data packet can be modeled as a weighted sum of two exponential functions, expressed as:

[0147]

[0148] In the formula, β 1 , β 2 , β 3 , β 4 represent model parameters.

[0149] Define the semantic successful decoding variable as Z i (t, e), Z i (t, e) = 1 indicates that the e-th semantic data packet of the terminal d i is successfully decoded in the t-th time slot, otherwise Z i (t, e) = 0. Define PAoS i (t, e) to represent the peak semantic age of the e-th semantic data packet of the terminal d i in the t-th time slot. The calculation formula is:

[0150]

[0151] In the formula, ω i (t, e) represents the importance degree of the semantic data packet, reflecting the importance degree of the service data characteristics corresponding to the semantics for the hierarchical collaborative control of the distribution network. In the parentheses, the first item δ i (t, e) represents the end-to-end delay, the second item represents the waiting delay caused by the failure of semantic decoding, O e refers to the index of the next successfully decoded data packet after the e-th data packet, and the third item Δ i (t, O e ) represents the semantic decoding delay.

[0152] Define the peak semantic age of the terminal d i in the t-th time slot as the maximum value among the peak semantic ages of all data packets, expressed as:

[0153] PAoS i PAoS(t) = max e {PAoS i (t, e)} (8)

[0154] S1.2: The present invention minimizes the peak semantic age of the distribution network by jointly optimizing the sensing frequency selection, terminal scheduling, and semantic compression ratio. Define the optimization variables as f = {f i (t)}, x = {x i (t)}, y = {y i (t)}, and the optimization problem is constructed as follows:

[0155]

[0156] In the formula, C 1 is the data acquisition frequency constraint; C 2 is the terminal scheduling constraint, indicating that an edge server can receive data from at most q terminals simultaneously; C 3 is the semantic compression ratio constraint.

[0157] S1.3: Based on the differences of the optimization entities, the present invention splits the original problem P1 into the edge-side terminal scheduling sub-problem SP1 and the joint optimization sub-problem SP2 of the end-side acquisition frequency selection and semantic compression ratio.

[0158] 2. Based on Top-N 2 and knowledge-statistics hybrid-driven fuzzy reinforcement learning integrated sensing and communication resource allocation algorithm

[0159] S2: The present invention proposes an integrated sensing and communication resource allocation algorithm based on Top-N 2 and knowledge-statistics hybrid-driven fuzzy reinforcement learning to solve the edge-side terminal scheduling sub-problem SP1 and the joint optimization sub-problem SP2 of the end-side acquisition frequency selection and semantic compression ratio. First, optimize the terminal scheduling through Top-N 2 ; secondly, through the knowledge-statistics hybrid-driven strategy, integrate the advantages of fuzzy learning and reinforcement learning to realize the joint optimization of the sensing frequency selection and the semantic compression ratio.

[0160] S2.1: Construct the terminal scheduling problem as a Markov decision process, and the key elements are introduced as follows.

[0161] 1) State space

[0162] At the t-th time slot, the state space of the edge server is defined as the set of the state spaces of each terminal, denoted as where the state space of terminal d i is G i (t) is the edge-side terminal d at the t-th time slot iData queue backlog.

[0163] 2) Action

[0164] At time slot t, the action space of the edge server is defined as the set of terminal scheduling decisions, denoted as

[0165]

[0166] 3) Cost function

[0167] The optimization problem SP1 is a minimization problem. The cost function at time slot t is defined as the optimization objective of SP1, that is

[0168] Since the edge server cannot grasp information such as the semantic data importance and channel state of each terminal in this time slot in advance when optimizing terminal scheduling, it is difficult to solve using traditional optimization methods based on global deterministic information. Therefore, the present invention proposes Top-N 2 , using the DQN network to mine and fit the cost of scheduling terminal d i , that is, the Q value, and minimizing the cost function by scheduling the q terminals with the smallest Q value. The implementation steps of Top-N 2 algorithm are as follows:

[0169] Step 1: The edge server inputs the numbers and state spaces of each terminal into the DQN network to obtain the Q value

[0170] Step 2: Based on the Q value, the terminals are sorted in ascending order and the q terminals with the smallest Q value are selected for scheduling, which can be expressed as:

[0171]

[0172] Step 3: At the end of the time slot, the edge server updates the queues Q i (t + 1), L i (t + 1), G i (t + 1), observes the PAoS and semantic importance of each terminal, and calculates the loss function, which can be expressed as:

[0173]

[0174] Step 4: The edge server updates the DQN network based on the loss function and the gradient descent method

[0175] S2.2: Considering the limited computing power of the terminal, the problem of jointly optimizing the data collection frequency and semantic compression ratio is modeled as an MAB problem, and a lightweight UCB algorithm is used to solve it. Each arm of the slot machine corresponds to a combination scheme of the data collection frequency and the semantic compression ratio. The UCB algorithm balances exploration and exploitation. On the one hand, it uses statistical information to select the arm with the best known empirical performance, and on the other hand, it explores new arms to obtain better rewards, so as to obtain the maximum reward. The knowledge-statistics hybrid-driven fuzzy reinforcement learning algorithm proposed in the present invention consists of 3 steps: hybrid-driven arm utility evaluation, action formulation, and statistical learning parameter update. The specific implementation process is as follows:

[0176] Step 1: Hybrid-driven arm utility evaluation

[0177] The utility of the arm consists of two parts. One is the fitness of the arm with the operating states of the electrical equipment (primary side) and the terminal (secondary side) constructed based on knowledge-driven fuzzy learning. The second is the upper confidence bound of the arm performance constructed based on statistics-driven reinforcement learning.

[0178] 1) Fitness calculation

[0179] The present invention constructs a six-layer fuzzy network to mine the fitness of the arm with the operating state.

[0180] The first layer is the state input layer, which consists of six neurons, corresponding to six elements in the terminal-side state space respectively, namely the average semantic importance of the data collected by the terminal d i in the (t-1)th time slot queue backlog Q i (t), the data to be retransmitted in this time slot L i (t), the terminal scheduling result x i (t), the data collection frequency action f i (t), the semantic compression ratio action y i (t).

[0181] The second layer is the membership degree calculation layer, which is used to calculate the membership degree of each state for different fuzzy levels. The membership degree γ i,j,k (t) reflects the degree to which the jth state of the terminal d i in the time slot t belongs to the kth fuzzy level, and the calculation formula is:

[0182]

[0183] In the formula, μ i,j,k and φ i,j,k are the central value and the diffusion parameter of the kth fuzzy level for the state j respectively. The membership degree calculation layer converts the specific state value into a fuzzy value for further processing.

[0184] The third layer is the ignition intensity calculation layer. The fuzzy rules of the ignition intensity are expressed as the product of the membership degrees of each state, and are expressed as:

[0185]

[0186] The fourth layer is the normalization layer, which is used to normalize the ignition intensity. The normalized ignition intensity is expressed as

[0187] The fifth layer is the fully connected layer, which is used to calculate the activation intensity of each fuzzy rule, and can be expressed as:

[0188]

[0189] In the formula, is the activation intensity of the fuzzy rule selecting the m-th acquisition frequency action and the n-th semantic compression action at the t-th time slot; v i,j (t) represents the weight of the neural network; b i,j (t) is the bias.

[0190] The sixth layer is the output layer, which is used to output the adaptation degree result between the electrical equipment and the terminal, and can be expressed as:

[0191]

[0192] In the formula, W is the number of fuzzy rules, which is equal to the product of the number of fuzzy levels of each state.

[0193] 2) Utility calculation

[0194] In order to balance the influence of knowledge-driven and statistics-driven, it is necessary to normalize the output adaptation degree result, which can be expressed as:

[0195]

[0196] In the formula, the parameter ξ 1 →ξ 0 +, ξ 0 →0+ normalizes the adaptation degree to between (0, 1). Based on the normalized adaptation degree the utility value is calculated as:

[0197]

[0198] In the formula, θ i,m,n (t) represents the empirical performance of the terminal d i selecting the m-th acquisition frequency and the n-th semantic compression ratio up to the t-th time slot. α i,m,n(t) represents the number of times the m-th acquisition frequency and the n-th semantic compression ratio are selected up to the t-th time slot.

[0199] Step 2: Action formulation

[0200] Define the action formulation variable as λ i,m,n (t). λ i,m,n (t) = 1 indicates that the terminal d i selects the m-th acquisition frequency and the n-th semantic compression ratio in the t-th time slot, otherwise λ i,m,n (t) = 0. The terminal selects the combination of the acquisition frequency and the semantic compression ratio with the minimum utility value (m * , n * ), which can be expressed as:

[0201]

[0202] Step 3: Update of statistical learning parameters

[0203] At the end of the time slot, the terminal updates the statistical learning parameters based on the PAoS information fed back by the edge server as follows:

[0204]

[0205] Embodiment 2:

[0206] As Figure 2 shown, the present invention proposes a 6G power semantic communication system based on integrated communication and sensing, including an end-side control layer and a side-edge control layer. Each layer is introduced as follows:

[0207] The end-side control layer includes power semantic communication IoT terminals and electrical devices such as distributed photovoltaic, charging piles, flexible loads, and energy storage batteries. The power semantic communication IoT terminals collect and monitor the operating environment and operating status of the electrical devices. Based on the power service characteristics, the power semantic communication IoT terminals extract and compress the semantics of the collected raw data, and upload the compressed data to the side-edge control layer.

[0208] The side-edge control layer consists of a 6G base station and a power semantic communication computing gateway. The power semantic communication computing gateway decodes and processes the semantic information received by the 6G base station, and provides data for power services such as island operation judgment, fault handling, precise load shedding, and regional autonomous control. The data processed by the side-edge control layer can also be further uploaded to the cloud layer to support cross-regional coordinated control, distributed photovoltaic output prediction, load prediction, panoramic monitoring, and main-distribution coordinated control.

[0209] The power semantic communication IoT terminal includes a data acquisition module, a semantic encoding module, a semantic compression module, a channel encoding module, a semantic importance recognition module, a communication module, a queue backlog awareness module, a communication-sensing integrated computing module, and a power supply module. The power semantic communication computing gateway includes a channel decoding module, a semantic decoding module, a communication module, a peak semantic age calculation module, a terminal scheduling module, a data storage module, a data processing module, and a power supply module.

[0210] The composition of the power semantic communication IoT terminal is as Figure 3 shown, and the introduction of each module is as follows:

[0211] Data acquisition module: This module is responsible for acquiring the operation environment and operation status data of electrical equipment.

[0212] Semantic encoding module: This module is responsible for extracting semantic information from the original acquired data based on service characteristics.

[0213] Semantic compression module: This module is responsible for compressing the extracted semantic information to reduce the amount of data required for transmission.

[0214] Channel encoding module: This module is responsible for improving the reliability of semantic information transmission in a complex channel environment by adding redundant information.

[0215] Semantic importance recognition module: This module is responsible for recognizing the semantic importance of the acquired data.

[0216] Communication module: This module is responsible for sending information such as acquired data and semantic importance to the power semantic communication computing gateway, and receiving information on whether decoding is successful and terminal scheduling information sent by the power semantic communication computing gateway;

[0217] Queue backlog awareness module: This module is responsible for calculating the backlog of the acquired data queue.

[0218] Communication-sensing integrated computing module: This module is responsible for optimizing the terminal sensing frequency selection and semantic compression ratio through knowledge-statistics hybrid-driven fuzzy reinforcement learning, and sending control information to the data acquisition module and the semantic compression module.

[0219] Power supply module: This module is responsible for supplying power to other modules.

[0220] The composition of the power semantic communication computing gateway is as Figure 4 shown, and the introduction of each module is as follows:

[0221] Channel decoding module: This module is responsible for receiving signals and performing symbol detection to recover the transmitted compressed semantic information.

[0222] Semantic decoding module: This module is responsible for converting the information recovered by the channel decoding module back into semantic concepts related to the task.

[0223] Communication module: This module is responsible for receiving the semantic information received by the 6G base station and sending the information on whether the decoding is successful and the terminal scheduling information to the power semantic communication Internet of Things terminals.

[0224] Peak semantic age calculation module: This module is responsible for calculating the peak semantic age of the received data.

[0225] Terminal scheduling module: This module is responsible for using the DQN network to mine and fit the cost of scheduling terminals, selecting appropriate terminals to minimize the terminal scheduling cost, and sending the terminal scheduling information to each power semantic communication Internet of Things terminal.

[0226] Data storage module: This module is responsible for storing the operation status data information of electrical equipment collected by the power semantic communication Internet of Things terminals.

[0227] Data processing module: This module is responsible for performing power operations such as islanding operation judgment, fault handling, precise load shedding, and regional autonomous control using the received operation status data information of electrical equipment.

[0228] Power supply module: This module is responsible for supplying power to other modules.

[0229] Embodiment 3:

[0230] Combining the 6G power semantic communication method based on integrated communication and sensing proposed in the above Embodiment 1 and the 6G power semantic communication system based on integrated communication and sensing proposed in the above Embodiment 2, it can be understood that:

[0231] The present invention proposes a new information timeliness metric that combines the peak information age and the semantic decoding success rate, namely the peak semantic age (PAoS); compared with the traditional information timeliness metric, PAoS not only covers the entire life cycle of information from information collection, semantic compression, semantic transmission to semantic parsing, but also particularly considers the successful parsing ability of the edge server for the received important semantic information. This measurement method further integrates the success rate of semantic information parsing on the basis of the traditional information age, enabling the metric to more comprehensively and accurately reflect the actual value and timeliness of information in the power system. By carefully considering the key performance indicators in the semantic compression and transmission process of information, the present invention can not only effectively guide the hierarchical collaborative control strategy of the distribution network, but also significantly improve the operation efficiency and reliability of the power system. Especially in the context of 6G communication technology, the PAoS metric of the present invention provides strong technical support for the efficient and intelligent management of the power system.

[0232] The present invention proposes a method based on Top-N 2The terminal scheduling optimization method accurately mines and fits the terminal scheduling cost through the application of the edge-side deep Q-network. This method combines fuzzy reinforcement learning to effectively jointly optimize the acquisition frequency and semantic compression ratio. It uses the DQN network to extract the features of each terminal state space and then generates the Q value as the decision basis. Greedy scheduling based on the Q-value ranking satisfies all terminals with quotas to minimize the average PAoS. During this process, the algorithm continuously updates the queue status of the terminals, the observed PAoS, and the semantic importance. At the end of each scheduling cycle, the system updates the DQN model based on the actual feedback, which includes the calculation of the loss function and the gradient descent optimization of the network parameters. The loss function reflects the difference between the actual PAoS and the PAoS predicted by the DQN, and its optimization is directly related to the improvement of the scheduling strategy accuracy and scheduling efficiency. In this process, the core of the algorithm is to finely balance resource allocation, dynamically adjust decisions according to the real-time network status and data requirements, aiming to optimize the response speed and data processing ability of the power system.

[0233] The present invention proposes a fuzzy reinforcement learning method based on a hybrid drive of knowledge and statistics, which uses the combination of a deep fuzzy neural network and the UCB algorithm to refine the mapping between the operation state and decision-making of the distribution network. First, the operation state and decision-making of the primary and secondary sides of the distribution network are analyzed through the deep fuzzy neural network, and these complex operation states and decisions are transformed into relevant knowledge models. Subsequently, the knowledge model is embedded into the reinforcement learning framework based on statistical drive, enhancing the adaptability and prediction accuracy of the model for the power grid operation situation. Under this framework, by calculating the fitness of each action and performing normalization processing, and evaluating the utility of each action plan, the present invention establishes a dynamic decision-making process. This process can adjust the data acquisition frequency and semantic compression ratio according to the real-time state of the power grid while ensuring the accuracy of the decision, optimizing the resource allocation strategy. Finally, by continuously updating the statistical learning parameters, this method can adapt to the changes in the operation of the distribution network, ensuring the continuous optimization and efficiency improvement of the decision-making process. This systematic methodology provides a new intelligent solution for the resource allocation and operation management problems involved in the hierarchical collaborative control of the power system.

[0234] The present invention proposes a 6G power semantic communication system device based on integrated sensing and communication. This system device covers the end-side control layer and the edge-side control layer, and each layer is equipped with targeted module designs to optimize the acquisition, processing, and transmission processes of power data. In the end-side control layer, through multiple modules of the Internet of Things terminal, such as data acquisition module, semantic encoding module, semantic compression module, etc., the present invention realizes the efficient monitoring of the operating state of electrical equipment and the accurate acquisition and compression of semantic information. These information are then uploaded to the edge-side control layer through the communication module and further processed by the channel decoding module and semantic decoding module of the power semantic communication edge computing gateway. The innovation of this system also includes the implementation of a knowledge-statistics hybrid-driven fuzzy reinforcement learning method. Through the integrated sensing and communication computing module and the terminal scheduling module, it realizes the intelligent optimization of the sensing frequency and semantic compression ratio, as well as the dynamic adjustment of the terminal scheduling strategy. This overall system device not only improves the timeliness and accuracy of information in the power system, but also optimizes resource allocation and improves the efficiency of power data processing and communication.

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

[0236] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device that implements in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0237] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0238] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

[0239] Many other changes and modifications can 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 6G power semantic communication method based on synaesthesia integration, characterized in that: The method comprises: Based on the constructed terminal-side perception model and communication model, a new information timeliness measurement indicator adapted to power semantic communication is constructed, namely the peak semantic age; Considering the constraints of distribution network hierarchical coordinated control on information timeliness, a joint optimization of perception frequency selection, terminal scheduling and semantic compression ratio is constructed to minimize the optimization problem of distribution network peak semantic age. Based on the differences of the optimization subjects, the original problem is split into the edge terminal scheduling sub-problem and the terminal acquisition frequency selection and semantic compression ratio joint optimization sub-problem; The terminal scheduling problem is constructed as a Markov decision process. 2 Optimize terminal scheduling; The joint optimization problem of data collection frequency and semantic compression ratio is modeled as a MAB problem, and a six-layer fuzzy network is constructed to mine the adaptability of the rocker arm and the operating state. The rocker arm utility is calculated based on the normalized fitness, and the terminal selects the combination of acquisition frequency and semantic compression ratio with the smallest utility value. The terminal updates the statistical learning parameters based on the peak semantic age information fed back by the edge server.

2. According to claim 1, a 6G power semantic communication method based on synaesthesia integration is characterized in that: The constructing of the terminal side perception model specifically includes: Assume there are I IoT terminals, and the terminal set is represented as Determine terminal d i The acquisition frequency in the tth time slot is f i (t), determine f i,max and f i,min are the upper and lower limits of the acquisition frequency respectively, and the value range of the acquisition frequency is discretized into M levels, expressed as Among them, the acquisition frequency of the mth file is expressed as 3. According to the 6G power semantic communication method based on synaesthesia integration according to claim 1, it is characterized in that: The construction of the communication model specifically includes: Determine the terminal scheduling variable as x i (t), x i (t) = 1 means that in the tth time slot, terminal d i is scheduled to transmit the compressed semantics to the edge server; assuming that in the tth time slot, terminal d i Scheduled to upload compressed semantic information; Determine terminal d i The semantic compression ratio at the tth time slot is the ratio of the amount of data after compression to the amount of data before compression, expressed as y i (t); determine y i,max and i,min are the upper and lower limits of the semantic compression ratio respectively. The value range of the semantic compression ratio is discretized into N levels, expressed as Among them, the semantic compression ratio of the nth level is expressed as Considering that the compressed semantics are quantized into bits and transmitted via 6G, at the tth time slot, from terminal d i The transmission rate to the edge server is: In the formula, B i For terminal d i The transmission bandwidth between the edge server and the i Yes i The transmission power, h i (t) is d i The channel gain between the edge server and the edge server, N0 is the noise power spectral density, N i (t) is d i The electromagnetic interference power; Terminal d i The collected data is cached locally to form a data queue. The backlog of the data queue is: Q i (t+1)=Q i (t)+f i (t)a i +L i (t)-U i (t)(2) In the formula, the input of the queue is Q i (t), f i (t)a i and L i (t); f i (t) represents the terminal d of the tth time slot i The acquisition frequency, a i is the amount of data collected in a single time, L i (t) is the amount of data that needs to be retransmitted due to semantic decoding errors, U i (t) is the amount of data transmitted, indicating the output of the queue, and is calculated as: Where the time slot length is τ, [·] - represents the minimum function, Represents terminal d i The semantic compression rate in the tth time slot is calculated as: In the formula, It is terminal d i The computing resources required to compress the semantic data per bit, It is terminal d i Computational resources used to perform semantic compression tasks.

4. According to the 6G power semantic communication method based on synaesthesia integration according to claim 1, it is characterized in that: The construction of new information timeliness measurement indicators specifically includes: Building a mapping function represents the terminal d processed by the edge server in the tth time slot i The e-th semantic data packet is transmitted in the t′th time slot, and y i (t) is terminal d i The semantic compression ratio at the tth time slot is, then the decoding delay of the eth semantic data packet is: In the formula, represents the computing resources required by the edge server to decode the unit bit of semantic data in time slot t, represents the computing resources used by the edge server to perform semantic decoding tasks; a i is the amount of data collected in a single time; Determine S i (t,e) is the decoding success rate of the e-th semantic data packet in the t-th time slot. The decoding success rate of the semantic data packet can be modeled as a weighted sum of two exponential functions, expressed as: In the formula, β1, β2, β3, and β4 represent model parameters; Determine the semantic success decoding variable as Z i (t,e), Z i (t,e)=1 indicates terminal d i The e-th semantic data packet of is successfully decoded in the t-th time slot, otherwise Z i (t,e)=0; Determine PAoS i (t,e) indicates terminal d i The peak semantic age of the e-th semantic data packet at the t-th time slot is calculated as: In the formula, ω i (t,e) represents the importance of the semantic data packet, reflecting the importance of the business data features corresponding to the semantics for the hierarchical coordinated regulation of the distribution network; in the brackets, the first term δ i (t,e) represents the end-to-end delay, the second term Indicates the waiting delay caused by semantic decoding failure, O e is the index of the next successfully decoded packet after the eth packet, and the third term Δ i (t,O e ) represents semantic decoding delay; Determine terminal d i The peak semantic age at the tth time slot is the maximum value of the peak semantic ages of all packets, expressed as: PAoS i (t)=max e {PAoS i (t,e)}(8) 5. According to the 6G power semantic communication method based on synaesthesia integration according to claim 1, it is characterized in that: By jointly optimizing the perception frequency selection, terminal scheduling and semantic compression ratio, the peak semantic age of the distribution network is minimized; the optimization variable is determined as f = {f i (t)}, x={x i (t)}, y={y i (t)}, the optimization problem is constructed as: In the formula, I is the number of IoT terminals, C1 is the data collection frequency constraint; C2 is the terminal scheduling constraint, which means that an edge server can receive data from up to q terminals at the same time; C3 is the semantic compression ratio constraint; PAoS i (t) indicates terminal d i The peak semantic age at the tth time slot is the maximum value among the peak semantic ages of all packets; is the semantic compression ratio set of the terminal, T means dividing the total optimization time into T time slots, is the terminal's collection frequency set; Based on the differences in the optimization subjects, the original problem P1 is split into the edge terminal scheduling sub-problem SP1 and the terminal acquisition frequency selection and semantic compression ratio joint optimization sub-problem SP2.

6. The 6G power semantic communication method based on synaesthesia integration according to claim 1 is characterized in that: The terminal scheduling problem is constructed as a Markov decision process. 2 Optimize terminal scheduling, including: At the tth time slot, the state space of the edge server is defined as the set of the state spaces of each terminal, expressed as Among them, terminal d i The state space of G i (t) is the edge terminal d in the tth time slot i The data queue is backlogged; Represents the computing resources used by edge servers to perform semantic decoding tasks; PAoS i (t) indicates terminal d i The peak semantic age at the tth time slot is the maximum value of the peak semantic ages of all packets; L i (t) is the amount of data that needs to be retransmitted due to semantic decoding errors; At the tth time slot, the action space of the edge server is defined as the terminal scheduling decision set, expressed as The optimization problem SP1 is a minimization problem. The cost function of the tth time slot is defined as the optimization objective of SP1, that is, Using DQN network to mine and fit scheduling terminal d i The cost, i.e., Q value, is minimized by scheduling the q terminals with the smallest Q value. 2 The implementation steps of the algorithm include: The edge server inputs each terminal number and its state space into the DQN network In the Q value Arrange the terminals in ascending order based on the Q value and select the q terminals with the smallest Q value for scheduling, which can be expressed as: At the end of the time slot, the edge server updates the queue Q i (t+1), L i (t+1), G i (t+1), observe the PAoS and semantic importance of each terminal, and calculate the loss function, which is expressed as: T represents the division of the total optimization time into T time slots, and I represents the set of IoT terminals. The edge server updates the DQN network based on the loss function and gradient descent method.

7. The 6G power semantic communication method based on synaesthesia integration according to claim 1 is characterized in that: Considering the limited computing power of the terminal, the joint optimization problem of data collection frequency and semantic compression ratio is modeled as a MAB problem and solved by a lightweight UCB algorithm. The rocker arm of each slot machine corresponds to a combination of data collection frequency and semantic compression ratio. The UCB algorithm balances exploration and utilization. On the one hand, it uses statistical information to select the rocker arm with the best performance known from experience, and on the other hand, it explores new rocker arms to obtain better benefits, thereby achieving the maximum benefit. The joint optimization problem of data collection frequency and semantic compression ratio is modeled as a MAB problem, and a six-layer fuzzy network is constructed to mine the adaptability of the rocker arm and the operating state, including three steps: hybrid drive rocker arm utility evaluation, action formulation, and statistical learning parameter update; The specific process includes: Step 1: Hybrid drive rocker arm effectiveness evaluation: The utility of the rocker arm includes two parts: one is the compatibility between the rocker arm and the operating state of the electrical equipment (i.e., the primary side) and the terminal (i.e., the secondary side) constructed based on knowledge-driven fuzzy learning; the second is the upper confidence bound of the rocker arm performance constructed based on statistics-driven reinforcement learning; 1) Calculation of fitness A six-layer fuzzy network is constructed to explore the adaptability of the rocker arm and the operating state; The first layer is the state input layer, which consists of six neurons corresponding to the terminal side state space The six elements in the t-1 time slot terminal d i Average semantic importance of collected data Queue backlog Q i (t), the data to be retransmitted in this time slot L i (t), terminal scheduling result x i (t), acquisition frequency action f i (t), semantic compression ratio action y i (t); The second layer is the membership calculation layer, which is used to calculate the membership of each state for different fuzzy levels. The membership γ i,j,k (t) reflects the terminal d i The degree to which the jth state at time slot t belongs to the kth fuzzy level is calculated as: In the formula, μ i,j,k and φ i,j,k They are the central value and diffusion parameter of the kth fuzzy level for state j. The membership calculation layer converts the specific state value into a fuzzy value; The third layer is the ignition intensity calculation layer, fuzzy rules The ignition intensity is expressed as the cumulative multiplication of the membership of each state, expressed as: The fourth layer is the normalization layer, which is used to normalize the ignition intensity. The normalized ignition intensity is expressed as The fifth layer is a fully connected layer, which is used to calculate the activation strength of each fuzzy rule, expressed as: In the formula, Fuzzy rules Select the activation strength of the mth acquisition frequency action and the nth semantic compression action in time slot t; v i,j (t) represents the weight of the neural network; b i,j (t) is the bias; The sixth layer is the output layer, which is used to output the compatibility results between the electrical equipment and the terminal, expressed as: Where W is the number of fuzzy rules, which is equal to the cumulative multiplication of the number of fuzzy levels of each state; 2) Utility calculation In order to balance the influence of knowledge-driven and statistical-driven, the output fitness results are normalized and expressed as: In the formula, the parameters ξ1→ξ0+,ξ0→0+ normalize the fitness to between (0,1); based on the normalized fitness The calculated utility value is: In the formula, θ i,m,n (t) indicates terminal d i The empirical performance of selecting the mth acquisition frequency and the nth semantic compression ratio by the tth time slot; α i,m,n (t) represents the number of times the mth acquisition frequency and the nth semantic compression rate are selected up to the tth time slot; Step 2: Action planning: Determine the action setting variable as λ i,m,n (t);λ i,m,n (t) = 1 indicates terminal d i Select the mth acquisition frequency and the nth semantic compression ratio in the tth time slot, otherwise λ i,m,n (t) = 0; the terminal selects the combination of acquisition frequency and semantic compression ratio with the smallest utility value (m * ,n * ), expressed as: Step 3: Statistical learning parameter update: At the end of the time slot, the terminal updates the statistical learning parameters based on the PAoS information fed back by the edge server: PAoS i (t) indicates terminal d i The peak semantic age at the tth time slot is the maximum value of the peak semantic ages of all packets.

8. A 6G power semantic communication system based on synaesthesia integration, characterized in that: The system includes: an edge layer and a terminal layer; the terminal layer includes a power semantic communication Internet of Things terminal and electrical equipment; the edge layer includes a 6G base station and a power semantic communication computing gateway; The electric power semantic communication Internet of Things terminal includes: a data acquisition module, a semantic coding module, a semantic compression module, a channel coding module, a semantic importance recognition module, a communication module, a queue backlog perception module, a synaesthesia integrated computing module and a power supply module; The data acquisition module is used to collect the operating environment and operating status data of the electrical equipment; The semantic encoding module is used to extract semantic information from the above-mentioned original collected data based on business characteristics; The semantic compression module is used to compress the extracted semantic information; The channel coding module is used to increase redundant information and improve the reliability of semantic information transmission in a complex channel environment; The semantic importance recognition module is used to recognize the semantic importance of the collected data; The communication module is used to send the collected data and semantic importance information to the power semantic communication computing gateway, and receive the information on whether the decoding is successful and the terminal scheduling information issued by the power semantic communication computing gateway; The queue backlog sensing module is used to calculate the backlog of the collection data queue; The synaesthesia integrated calculation module is used to optimize the terminal perception frequency selection and semantic compression ratio through knowledge-statistics hybrid driven fuzzy reinforcement learning, and send control information to the data acquisition module and the semantic compression module; The power module supplies power to the data acquisition module, the semantic coding module, the semantic compression module, the channel coding module, the semantic importance recognition module, the communication module, the queue backlog perception module and the synaesthesia integrated computing module; The electric power semantic communication computing gateway comprises: a channel decoding module, a semantic decoding module, a communication module, a peak semantic age calculation module, a terminal scheduling module, a data storage module, a data processing module and a power supply module; The channel decoding module is used to receive the signal and perform symbol detection to recover the transmitted compressed semantic information; The semantic decoding module is used to convert the information recovered by the channel decoding module back into semantic concepts related to the task; The communication module is used to receive the semantic information received by the 6G base station, and send information on whether the decoding is successful and terminal scheduling information to the power semantic communication Internet of Things terminal; The peak semantic age calculation module is used to calculate the peak semantic age of the received data; The terminal scheduling module is used to use the DQN network to mine and fit the cost of scheduling terminals, select appropriate terminals to minimize terminal scheduling costs, and send terminal scheduling information to each power semantic communication Internet of Things terminal; The data storage module is used to store the electrical equipment operating status data information collected by the power semantic communication Internet of Things terminal; The data processing module is used to use the received electrical equipment operation status data information to perform island operation analysis, fault handling, precise load shedding and regional autonomous control of power services; The power supply module supplies power to the channel decoding module, the semantic decoding module, the communication module, the peak semantic age calculation module, the terminal scheduling module, the data storage module and the data processing module.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method according to any one of claims 1 to 7.

Images