CN118338321A - Electric power semantic short packet communication method, device and system based on general sense integration - Google Patents
Electric power semantic short packet communication method, device and system based on general sense integration Download PDFInfo
- Publication number
- CN118338321A CN118338321A CN202410516599.8A CN202410516599A CN118338321A CN 118338321 A CN118338321 A CN 118338321A CN 202410516599 A CN202410516599 A CN 202410516599A CN 118338321 A CN118338321 A CN 118338321A
- Authority
- CN
- China
- Prior art keywords
- semantic
- power
- terminal
- data
- paois
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004891 communication Methods 0.000 title claims abstract description 139
- 238000000034 method Methods 0.000 title claims abstract description 106
- 230000010354 integration Effects 0.000 title claims abstract description 35
- 238000005457 optimization Methods 0.000 claims abstract description 111
- 230000005540 biological transmission Effects 0.000 claims abstract description 31
- 230000008447 perception Effects 0.000 claims abstract description 18
- 230000009471 action Effects 0.000 claims description 40
- 230000006835 compression Effects 0.000 claims description 40
- 238000007906 compression Methods 0.000 claims description 40
- 230000006870 function Effects 0.000 claims description 34
- 238000004422 calculation algorithm Methods 0.000 claims description 28
- 238000005259 measurement Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 19
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000005070 sampling Methods 0.000 claims description 9
- 230000007704 transition Effects 0.000 claims description 9
- 238000004458 analytical method Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 238000003745 diagnosis Methods 0.000 claims description 7
- 230000001052 transient effect Effects 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000012544 monitoring process Methods 0.000 claims description 6
- 230000035772 mutation Effects 0.000 claims description 6
- 238000013500 data storage Methods 0.000 claims description 5
- 230000007774 longterm Effects 0.000 claims description 5
- 230000007423 decrease Effects 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000005065 mining Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000006641 stabilisation Effects 0.000 claims description 3
- 238000011105 stabilization Methods 0.000 claims description 3
- 230000006872 improvement Effects 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000007246 mechanism Effects 0.000 abstract description 2
- 239000007787 solid Substances 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000007786 learning performance Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013468 resource allocation Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000001360 synchronised effect Effects 0.000 description 1
Abstract
The invention relates to a power semantic short packet communication method, device and system based on general sense integration, and belongs to the technical field of power communication. According to the invention, through constructing the multi-mode experience playback pool and introducing the electric power mode exploration-utilization compromise coefficient, the PMU terminal can dynamically adjust the data acquisition frequency, effectively cope with the perception challenges under the diversified operation modes of the power distribution network, improve the optimization performance of the perception frequency under the sparse mode, and strengthen the capturing capability of the key operation state of the electric power system. The method has the advantages that the uncertainty environment is simulated by the counteragent, the cooperative game mechanism of multiple agents is combined, the dynamic optimization and the robustness improvement of the information source channel joint coding are realized, the transmission reliability of the electric power semantic information under the complex channel condition is effectively ensured, and a solid technical support is provided for the efficient and reliable operation of an electric power system.
Description
Technical Field
The invention relates to a power semantic short packet communication method, device and system based on general sense integration, and belongs to the technical field of power communication.
Background
The uncertainty and intermittence of the output of the massive distributed new energy sources and the large-scale irregular charging of the electric automobile have great influence on the safe and reliable operation of the power distribution network, and real-time sensing of the operation state of the power distribution network is needed. Synchronous phasor measurement device (Phasor Measurement Unit, PMU) of distribution network possesses high accuracy measurement ability, can fast, accurately catch disturbance problems such as distribution network voltage sag, flicker, frequency mutation, supports second level accurate state perception. On the other hand, high-precision and millisecond-level massive PMU measurement data are required to be higher in real-time sensing and reliable transmission capacity of the 6G wireless communication network. Semantic communication realizes the conversion from traditional symbol transmission to a new semantic communication paradigm of 'understanding first and transmitting later', by extracting semantic features of original data and taking semantics as basic units of information characterization, and provides a solution for transmitting massive PMU measurement data by using 6G. Considering the short packet characteristics of PMU measurement data, the power semantic short packet communication is further combined with the general sense integration, under the limited block length normal form, the semantic compression and the information source channel joint coding are cooperatively optimized based on the power service characteristics, and the data transmission redundancy is reduced while the second-level accurate state sensing of the power distribution network is ensured. However, considering the multi-scale coupling of sensing and communication and the complex evolution of the power operation mode, how to design a power semantic short packet communication method, device and system based on sense of general integration, and by integrating the device adopting the method into the system, the joint optimization of sensing frequency, semantic compression ratio, semantic information merging and encoding quantity and data packet length is realized, so that the high timeliness reliable transmission of the key semantic information of the power service is still a core technical challenge.
In view of the above-mentioned drawbacks, the present invention is to create a method, apparatus and system for communication of short packets of power semantics based on sense of general integration, which makes the method, apparatus and system have more industrial utilization value.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a power semantic short packet communication method, device and system based on sense of general integration.
The invention discloses a power semantic short packet communication method based on general sense integration, which comprises the following specific steps:
Constructing a power semantic short packet communication system model based on communication integration, wherein the power semantic short packet communication system model comprises a PMU measurement data sensing model, a source channel joint coding model, a source channel joint decoding model and a peak semantic error age model facing power short packet communication;
Constructing a multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem oriented to power semantic short packet communication based on the power semantic short packet communication system model, and converting the multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem oriented to power semantic short packet communication into a single-time-slot optimization problem through decoupling among time slots;
The large-scale sensing frequency optimization algorithm driven by the electric power in a multi-mode and the small-scale information source channel joint coding robust optimization algorithm based on the cooperative game DAC are provided, a single-time slot optimization problem is solved, and the information source channel joint coding of accurate sensing of the running situation and electric power semantic short packet communication is realized.
Further, the PMU measured data sensing model specifically includes:
Defining the measured data sensing frequency of the PMU terminal d n in the period i as f n (i), discretizing the value of f n (i) into M-grade, and representing as:
wherein F n,max and F n,min are respectively the upper and lower limits of the sensing frequency, and the m-th gear sensing frequency is The total optimized time length is divided into I large time scales, namely time periods, and the aggregate isEach period is composed of T 0 small time scale time slots, wherein the time slot set of the ith period isThe total number of slots is t=it 0; define the number of PMU terminals as N and aggregate as
Further, the source channel joint coding model specifically includes:
Defining the semantic compression ratio of the terminal d n at the t time slot as x n (t), discretizing the value into K grade, and representing as Wherein x n,max and x n,min are respectively the upper and lower limits of the semantic compression ratio, and the semantic compression ratio of the kth gear is expressed as
Defining the valid information bit of the data perceived by d n each time after semantic coding as x n(t)An, wherein A n is the original data size; in channel coding, defining a packet length variable as l n (t); defining the variable of the combined coding quantity of the semantic information as y n(t)＝{1,2,…,j,...yn,max},yn (t) =j to represent that the time slot d n carries out joint channel coding on j semantic information; under the normal form of the joint coding of the information source channel and the finite block length, in order to ensure low transmission delay, the data packet length of t time slots needs to meet the following constraint:
ln(t)＞xn(t)yn(t)An (2)
Therefore, the transmission delay of the data packet with the length l_ { n } (t) is:
in the method, in the process of the invention, The channel decoding error probability for the data packet of the tth slot terminal d n, R n(γn,t,ln (t),The maximum channel coding rate at the t-th slot for terminal d n is expressed as:
Where γ n,t is the signal-to-noise ratio, R SC(γn,t) is d n and the shannon channel capacity of the edge computation gateway, V (γ n,t) is the channel dispersion, expressed as 1- (1 γ n,t)-1; Is that Is used as a reference to the remainder of the (c),As a Gaussian functionIs an inverse function of (c).
Further, the source channel joint decoding model specifically includes:
defining the channel decoding error probability of the data packet of the terminal d_n of the t time slot as follows:
Constructing a mapping relation function omega n,t,j,e = t ', representing the j-th semantic in the e-th data packet decoded by the edge computing gateway in the t-th time slot, and carrying out semantic coding by the terminal d n in the t' time slot; the semantic decoding error probability is:
where β 1、β2、β3、β4 denotes the model parameters, which can be learned by the Levenberg-Marquardt method, and which vary with the neural network employed.
Further, the peak semantic error age model for power short packet communication specifically includes:
definition τ n,t,e,j, Delta n,t,e,j represents the collection time, the semantic decoding completion time and the time interval between the data corresponding to the j-th semantic in the e-th data packet of the terminal d n decoded by the t-th time slot and the next semantic decoding completion time of the edge computing gateway respectively; definition PAoI n,t,e,j represents the jth semantic PAoI of the edge computing gateway in the ith data packet of the jth slot decoding terminal d n, expressed as:
in the method, in the process of the invention, Representing end-to-end delay; thus, PAoI of terminal d n is:
defining an innovative information timeliness measurement index suitable for power short packet communication, namely PAoIS is as follows:
wherein ω n (t, e, j) represents the importance of the j-th semantic in the e-th packet of the terminal d n; Representing decoding semantics With true semanticsDeviation exists between the two, namely semantic decoding errors; σ n denotes semantic decoding error weights;
the maximum PAoIS obtained by traversing all data packets and semantics is defined as PAoIS of the terminal d n on the edge computing gateway at the t-th time slot, and is expressed as:
constructing epsilon n -PAoIS constraint based on the law of large deviation, ensuring that the probability of the terminal PAoIS exceeding the threshold value is exponentially reduced, which is expressed as:
Wherein PAoIS n,th represents the PAoIS threshold value of terminal d n, ε n is PAoIS the decay rate of the boundary failure probability, and represents the probability that PAoIS of terminal d n exceeds threshold PAoIS n,th as the threshold value increases Descending;
in addition, the second-level accurate state sensing of the power distribution network requires that measurement data of all PMU terminals are on the same information timeliness section, and PAoI deviation constraint is met, which is expressed as:
Where ε' represents the maximum PAoI deviation of the accurate state sensing requirement of the power distribution network.
Further, the multi-time scale heterogeneous resource high timeliness collaborative optimization problem facing the power semantic short packet communication is modeled as follows:
Wherein, C 1 is the constraint of the sensing frequency, the semantic compression ratio and the semantic information merging coding quantity; c 2 is packet length constraint; c 3 is the ε n -PAoIS constraint; c 4 is PAoI bias constraint; the optimization variable is defined as f= { f n(i)}、x＝{xn(t)}、y＝{yn(t)}、l＝{ln (t) }.
Further, the single-slot optimization problem is specifically:
Introducing auxiliary variable H n (t) to record the condition that PAoIS exceeds a threshold value in the t time slot, and taking the condition as an optimization penalty, wherein the update rule is as follows
In the formula, the indication functionFor monitoring PAoIS n (t) if the threshold PAoIS n,th is exceeded; if the threshold value is exceeded, the function value is 1, otherwise, the function value is 0; exp (- ε nPAoISn,th) represents a forward penalty increment based on a threshold, decreases with increasing threshold in the update of H n (t+1), and represents a relative increase in the tolerance of the accurate state sensing of the power distribution network to conditions exceeding the threshold; based on the auxiliary variable H n (t), the optimization objective of P1 can be converted to ψ n (t), expressed as
Wherein the optimization objective ψ n (t) further considers the additional penalty caused by exceeding the threshold on the basis of PAoIS; thus, by inter-slot decoupling, the long-term optimization problem P1 is decomposed into a single-slot optimization problem P2, denoted as
Further, the large-scale sensing frequency optimization algorithm of the electric power multi-mode driving specifically comprises the following steps:
Based on PMU measurement data, mining multi-mode operation characteristics of the power distribution network, including Z modes such as voltage sag, flicker, frequency mutation and the like; defining a mode indicating variable as alpha n (t), wherein alpha n (t) =z represents that the measured data perceived by the terminal d n in the t time slot corresponds to the z-th operation mode; definition of the definition AndAverage PAoIS, average PAoI and average cost function for terminal d n during period (i-1); the state space of the large-scale perception frequency optimization problem is that
The action space of the terminal d n isConstructing a cost function of d n as
(1) DQN network initialization and multi-modal experience playback pool construction
Each terminal initializes its DQN model, including a main networkFor generating data-aware frequency decisions, a target networkFor stabilization training; constructing Z experience playback sub-pools respectively according to Z power distribution network modes, wherein the Z experience playback sub-pools are used for storing experience samples of different modes, and the aggregate of the experience samples is denoted as gamma n(i)＝{Υn,1(i),…,Υn,z(i),...,Υn,Z (i);
(2) Perceptual frequency decision and multi-modal experience updating
In the ith large-scale period, the main network is based on an E-greedy strategy and a Q valueSelecting an optimal actionAt the end of the period of time,Transition to the next stateTerminal calculation cost functionFinally, constructing an experience sampleAnd based on alpha n (t), storing the data into an experience playback sub-pool of the corresponding mode;
(3) Multimode driven DQN network optimization
Defining a power modality exploration-utilizing a trade-off factor κ n (i) for adjusting sampling strategies, the larger κ n represents the more exploration prone; in the ith large scale period, the empirical importance weight φ n,z (i) of the z-th modal empirical playback sub-pool of terminal d n is represented as
Where TD n,z,o represents the TD error for the o-th experience, |y n,z (i) | represents the number of experiences in the z-th experience playback sub-pool, and ||y n (i) | represents the total number of experiences in all experience playback pools;
Extracting a playback experience sample set phi n (i) by using a weighted sampling method and based on importance weight, and performing network training based on the playback experience sample set phi n (i); the loss function xi n (i) is calculated as
In the method, in the process of the invention,Terminal d n representing target network evaluation in stateThe minimum value of the Q values corresponding to all the next possible actions f n (i'),Is a discount factor;
Finally, the terminal d n optimizes the main network based on the gradient descent method and the loss function xi n (i) Every I 0 large time scales, updating the target network to be
Further, the small-scale information source channel joint coding robust optimization algorithm based on the cooperative game DAC specifically comprises the following steps:
(1) Anti-partnership POMDP construction
The problem of joint coding optimization of small-scale source channels is built into an anti-cooperation POMDP model, and the tuple is defined as
In the method, in the process of the invention,And (3) withRepresenting the state space, the set of action spaces, respectively, P ini is the conditional transition probability between the initial states,AndAn uncertainty set representing all state transition probabilities and cost functions; And Respectively, a part of observable states and conditional probabilities of the observable states; η is a discount factor;
(2) Algorithm execution flow
The proposed algorithm comprises an offline centralized learning phase and an online distributed learning phase:
1) Offline centralized learning phase
Step one: defining an anti-agent, an offline action network, and a joint evaluation network asAndThe edge computing gateway extracts H groups of experiences from the experience pool to perform offline training; wherein the decision of the antibody intelligent network and the action network based on the experience of the terminal d n in the h group is expressed as
In the method, in the process of the invention,Representing the observation state space of the terminal d n in the h group of experience, Δγ n,h representing the actions of the anti-agent on the anti-agent network in the offline centralized learning phase;
Step two: in the h sample, the combined consideration PAoI and PAoIS builds an action network with a challenge-co cost function of
Step three: TD errors of constructing action network, countermeasure network and joint evaluation network are as follows
In the method, in the process of the invention,Representing a system state after the action is performed;
Step four: based on TD error, training and updating action network by gradient descent method And an anti-agent networkRepresented as
Joint evaluation networkTraining updates are expressed as
2) Online distributed learning phase
Step one: each terminal uses the off-line training action network issued by the edge computing gateway as a local on-line action network, namelyAnd based on this, make online optimization decisions, denoted as
Step two: observing state change of each terminal, and constructing an experience sampleAnd placing the model into an experience pool; wherein,Indicating that terminal d n is policy basedAnd performing the optimal action of joint coding of the source channels.
10. Electric power semantic short packet communication system based on sense integration, characterized in that: the device comprises a terminal layer and an edge layer;
The terminal layer comprises a power semantic short packet communication PMU terminal, and the power semantic short packet communication PMU terminal comprises:
And a data acquisition module: the method comprises the steps of collecting operation state data of electrical equipment;
Semantic coding module: the method comprises the steps of preprocessing and encoding collected original data based on service characteristics, and extracting key information, namely semantics, in the data;
the semantic compression module: the method is used for compressing the encoded semantic information so as to reduce the bandwidth required by data transmission;
And the information source channel joint coding module: the method is used for combining source coding, namely semantic coding and compression, and channel coding, adding redundant information to resist noise and interference in a channel, and improving the reliability of data transmission;
And a communication module: the method comprises the steps of sending information such as semantic compression ratio, semantic information merging coding quantity, data packet length and the like to a power semantic short packet communication edge computing gateway, and receiving information such as maximum PAoIS, average PAoIS, average PAoI, PAoIS boundary damage probability attenuation rate and the like issued by the power semantic short packet communication edge computing gateway;
And the general sense integrated calculation module is as follows: the system is used for analyzing and processing the acquired data in real time, supporting the integrated optimization of sensing and communication and dynamically adjusting the acquisition frequency;
and a power supply module: for providing stable power support for the terminal device;
The edge layer comprises a 6G base station and an electric power semantic short packet communication edge computing gateway; the power semantic short packet communication edge computing gateway comprises:
and a channel decoding module: the PMU terminal is used for processing the data packet received from the PMU terminal, carrying out signal detection and channel decoding, recovering compressed semantic information and ensuring the integrity and reliability of data;
semantic decoding module: the system comprises a channel decoding module, a power system operation state generation module, a compression semantic information generation module and a power system operation state generation module, wherein the channel decoding module is used for receiving compressed semantic information output by the channel decoding module and converting the compressed semantic information back to original semantics related to the power system operation state; the step involves analyzing semantic codes, reconstructing key information of original data, and providing a basis for subsequent data processing and analysis;
And a communication module: the PMU terminal is used for exchanging data with the PMU terminal; supporting high-speed data receiving and sending through a 6G network, sending information such as maximum PAoIS and average PAoIS, average PAoI, PAoIS boundary failure probability attenuation rate and the like to a power semantic short packet communication PMU terminal, and receiving information such as semantic compression ratio, semantic information merging coding quantity, data packet length and the like sent by the power semantic short packet communication PMU terminal;
Peak semantic error age calculation module: PAoIS for calculating each data packet based on the received data, and evaluating the timeliness and accuracy of the data; the module is important to ensure that the data processing meets the accurate state sensing requirement of the power distribution network;
And a data storage module: the power distribution network key circuit transient state data information acquisition module is used for storing power semantic short packet communication PMU terminal acquired power distribution network key circuit transient state data information and providing support for state analysis and fault diagnosis;
And a data processing module: the method is used for further analyzing and processing the decoded data and providing data support for second-level accurate state sensing and operation trend early warning and fault diagnosis positioning of the power distribution network;
And a power supply module: the system is used for providing stable and continuous power supply for the power semantic short packet communication edge computing gateway, so that the system can run uninterruptedly;
A power semantic short packet communication device based on general sense integration comprises
Modeling module: the method is used for constructing a power semantic short packet communication system model based on communication sense integration, and comprises a PMU measurement data sensing model, a source channel joint coding model, a source channel joint decoding model and a peak semantic error age model facing power short packet communication;
And a conversion module: the method is used for constructing a multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem for power semantic short packet communication based on the power semantic short packet communication system model, and converting the multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem for power semantic short packet communication into a single-time-slot optimization problem through time-slot decoupling;
And a solving module: the method is used for solving a single-time-slot optimization problem by providing a large-scale perception frequency optimization algorithm driven by electric power multi-mode and a small-scale information source channel joint coding robust optimization algorithm based on a cooperative game DAC, and realizing information source channel joint coding of accurate perception of running situation and electric power semantic short packet communication.
Furthermore, in the conversion module, the multi-time scale heterogeneous resource high timeliness collaborative optimization problem facing the power semantic short packet communication is modeled as follows:
Wherein, C 1 is the constraint of the sensing frequency, the semantic compression ratio and the semantic information merging coding quantity; c 2 is packet length constraint; c 3 is the ε n -PAoIS constraint; c 4 is PAoI bias constraint; the optimization variable is defined as f= { f n(i)}、x＝{xn(t)}、y＝{yn(t)}、l＝{ln (t) }.
Further, in the conversion module, the single-slot optimization problem is specifically:
Introducing auxiliary variable H n (t) to record the condition that PAoIS exceeds a threshold value in the t time slot, and taking the condition as an optimization penalty, wherein the update rule is as follows
In the formula, the indication functionFor monitoring PAoIS n (t) if the threshold PAoIS n,th is exceeded; if the threshold value is exceeded, the function value is 1, otherwise, the function value is 0; exp (- ε nPAoISn,th) represents a forward penalty increment based on a threshold, decreases with increasing threshold in the update of H n (t+1), and represents a relative increase in the tolerance of the accurate state sensing of the power distribution network to conditions exceeding the threshold; based on the auxiliary variable H n (t), the optimization objective of P1 can be converted to ψ n (t), expressed as
Wherein the optimization objective ψ n (t) further considers the additional penalty caused by exceeding the threshold on the basis of PAoIS; thus, by inter-slot decoupling, the long-term optimization problem P1 is decomposed into a single-slot optimization problem P2, denoted as
By means of the scheme, the invention has at least the following advantages:
1. The peak error semantic age (PAoIS) measurement index provided by the invention creatively fuses information timeliness, semantic importance and decoding deviation consideration, accurately measures the full-link efficiency of power data transmission according to the real-time monitoring requirement of the power distribution network, and remarkably optimizes the accuracy and timeliness of power distribution network state perception.
2. The large-scale sensing frequency optimization algorithm driven by the electric power multi-mode enables the PMU terminal to dynamically adjust the data acquisition frequency by constructing a multi-mode experience playback pool and introducing electric power mode exploration-utilization compromise coefficients, effectively aims at sensing challenges under the diversified operation modes of the power distribution network, improves the optimization performance of the sensing frequency under the sparse mode, and strengthens the capturing capability of key operation states of the electric power system.
3. The small-scale information source channel joint coding robust optimization method based on the cooperative game DAC utilizes the anti-intelligent agent simulation uncertainty environment and combines the cooperative game mechanism of multiple intelligent agents to realize the dynamic optimization and the robust promotion of the information source channel joint coding, effectively ensures the transmission reliability of the electric power semantic information under the complex channel condition, and provides a solid technical support for the efficient and reliable operation of an electric power system.
The foregoing description is only an overview of the present invention, and is intended to provide a better understanding of the present invention, as it is embodied in the following description, with reference to the preferred embodiments of the present invention and the accompanying drawings.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate a certain embodiment of the present invention and therefore should not be considered as limiting the scope, and that other related drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a power semantic short packet communication system based on sense of general integration;
FIG. 2 is a schematic diagram of the PMU terminal composition for power semantic short packet communication according to the present invention;
FIG. 3 is a schematic diagram of the power semantic short packet communication edge computation gateway module of the present invention;
fig. 4 is a flow chart of a power semantic short packet communication method based on sense of general integration.
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
The invention comprises a power semantic short packet communication system and a power semantic short packet communication method based on sense of general integration, and the power semantic short packet communication system and the power semantic short packet communication method are specifically described as follows.
The invention provides an electric power semantic short packet communication system based on sense of general integration, which is shown in figure 1, and comprises a terminal layer and an edge layer, wherein the description of each layer is as follows:
The terminal layer mainly comprises a power semantic short packet communication PMU terminal. The PMU terminal is deployed on key lines of distribution networks such as distributed photovoltaic incoming and outgoing lines, and is used for measuring transient data such as voltage, current, frequency, power factor and the like in a high-precision millisecond level. The collected PMU measurement data is uploaded to an edge layer through an electric power semantic short packet communication Internet of things terminal through semantic extraction, semantic compression and channel coding. The edge layer consists of a 6G base station and an electric power semantic short packet communication edge computing gateway, wherein the electric power semantic short packet communication edge computing gateway performs channel decoding, semantic decoding and data processing on semantic information received by the base station, and provides data support for second-level accurate state sensing and operation trend early warning and fault diagnosis positioning of the power distribution network.
The power semantic short packet communication PMU terminal comprises a data acquisition module, a semantic coding module, a semantic compression module, a source channel joint coding module, a communication module, a sense-of-general integrated calculation module and a power module. The electric power semantic short packet communication edge computing gateway comprises a channel decoding module, a semantic decoding module, a communication module, a peak semantic error age computing module, a frequency acquisition module, a data storage module, a data processing module and a power supply module.
The power semantic short packet communication PMU terminal is shown in fig. 2, and the following modules are introduced:
And a data acquisition module: the method is used for collecting the operation state data of the electrical equipment.
Semantic coding module: the method is used for preprocessing and encoding the collected original data based on the service characteristics and extracting key information, namely 'semantics', in the data.
The semantic compression module: the method is used for compressing the encoded semantic information to reduce the bandwidth required during data transmission.
And the information source channel joint coding module: for combining source coding (i.e. semantic coding and compression) and channel coding, adding redundant information to combat noise and interference in the channel, improving the reliability of data transmission.
And a communication module: the method is used for sending the information such as semantic compression ratio, semantic information merging coding quantity, data packet length and the like to the electric power semantic short packet communication edge computing gateway and receiving the information such as maximum PAoIS, average PAoIS, average PAoI, PAoIS boundary damage probability attenuation rate and the like issued by the electric power semantic short packet communication edge computing gateway.
And the general sense integrated calculation module is as follows: the method is used for analyzing and processing the acquired data in real time, supporting the integrated optimization of sensing and communication and dynamically adjusting the acquisition frequency.
And a power supply module: for providing stable power support for the terminal device.
The power semantic short packet communication edge computing gateway module is shown in fig. 3, and the following description is provided for each module:
and a channel decoding module: the PMU terminal is used for processing the data packet received from the PMU terminal, carrying out signal detection and channel decoding, recovering compressed semantic information and ensuring the integrity and reliability of data.
Semantic decoding module: the compressed semantic information is used for receiving the compressed semantic information output by the channel decoding module and converting the compressed semantic information back to the original semantic related to the running state of the power system. This step involves parsing the semantic code, reconstructing the critical information of the original data, and providing the basis for subsequent data processing and analysis.
And a communication module: for data exchange with the PMU terminal. And supporting high-speed data receiving and sending through a 6G network, sending information such as maximum PAoIS and average PAoIS, average PAoI, PAoIS boundary failure probability attenuation rate and the like to the power semantic short packet communication PMU terminal, and receiving information such as semantic compression ratio, semantic information merging coding number, data packet length and the like sent by the power semantic short packet communication PMU terminal.
Peak semantic error age calculation module: PAoIS for computing each data packet based on the received data, and evaluating the timeliness and accuracy of the data. This module is crucial to guaranteeing that data processing satisfies the accurate state perception demand of distribution network.
And a data storage module: the power distribution network key circuit transient state data storage method is used for storing power semantic short packet communication PMU terminal collected power distribution network key circuit transient state data information and providing support for state analysis and fault diagnosis.
And a data processing module: the method is used for further analyzing and processing the decoded data and providing data support for second-level accurate state sensing and operation trend early warning and fault diagnosis positioning of the power distribution network.
And a power supply module: for providing a stable and continuous power supply for a power semantic short packet communication edge computing gateway, ensuring that the system can operate uninterrupted.
The invention provides a power semantic short packet communication method based on sense of general integration, and the specific flow is shown in figure 4.
S1: and constructing a power semantic short packet communication system model based on communication sense integration, wherein the power semantic short packet communication system model comprises a PMU measurement data sensing model, a source channel joint coding model, a source channel joint decoding model and a peak semantic error age model facing power short packet communication.
S2: the multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem oriented to power semantic short packet communication is constructed by jointly optimizing the sensing frequency, the semantic compression ratio, the semantic information merging coding quantity and the data packet length, so that PAoIS is minimized;
S3: the large-scale sensing frequency optimization algorithm driven by the electric power in a multi-mode and the small-scale information source channel joint coding robust optimization algorithm based on the cooperative game DAC are provided, so that the information source channel joint coding of accurate sensing of the running situation and electric power semantic short packet communication is realized.
In a further embodiment, step S1 comprises:
The invention provides an electric power semantic short packet communication system model based on general sense integration. Firstly, considering the 6G sense integrated resource allocation requirement facing to the accurate sensing of a power distribution network, respectively constructing a PMU measurement data sensing model, a power semantic short packet communication source channel joint coding model and a power semantic short packet communication source channel joint decoding model; secondly, constructing a peak error semantic age measurement index suitable for power semantic short packet communication, namely peak error semantic age, and covering all links of data acquisition, semantic coding, short packet transmission and semantic decoding; and finally, constructing a power semantic short packet communication information source channel joint coding optimization problem, and realizing high-timeliness reliable transmission of the multi-mode operation key characteristic information of the power distribution network by jointly optimizing the sensing frequency, the semantic compression ratio, the semantic information merging coding quantity and the data packet length and minimizing PAoIS. The concrete introduction is as follows:
S1.1: considering the scale difference between the running state change and the channel state change of the power distribution network, dividing the total optimization time length into I large time scales, namely time periods, and integrating the time periods into Each period is composed of T 0 small time scale time slots, wherein the time slot set of the ith period isThus, the total number of slots is t=it 0. Define the number of PMU terminals as N and aggregate asIn each period, the terminal optimizes the large time scale data sensing frequency, captures multimode operation state changes such as voltage sag, flicker, frequency mutation and the like of the power distribution network, and realizes accurate sensing of operation situation; in each time slot, the terminal optimizes the small time scale semantic compression ratio, the semantic information merging coding quantity and the data packet length, so that the source channel joint coding of the power semantic short packet communication is realized, and the high-timeliness and reliable transmission of the power semantic information is ensured.
S1.1.1: PMU measurement data perception model
The measured data sensing frequency of PMU terminal d n during period i is defined as f n (i). For simplicity, the value of f n (i) is discretized into M-range, expressed as:
wherein F n,max and F n,min are respectively the upper and lower limits of the sensing frequency, and the m-th gear sensing frequency is
S1.1.2: information source channel joint coding model for electric power semantic short packet communication
The power semantic short packet transmitting terminal is a PMU terminal and consists of a semantic encoder, a semantic compressor and a channel encoder. The semantic encoder extracts semantic information from the PMU-aware measurement data based on the differentiated features of the power traffic and the local semantic knowledge base. The semantic compressor further compresses the extracted semantic information based on the set semantic compression ratio, and data transmission redundancy is reduced. The channel encoder combines the coding quantity and the data packet length based on the set semantic information, codes the channel by adding redundant information, and improves the reliability of the semantic information transmission in a complex channel environment. The edge computing gateway is used as an electric power semantic short packet receiving end and is composed of a channel decoder and a semantic decoder. The channel decoder is used for receiving the signal and carrying out symbol detection to recover the transmitted compressed semantic information. The semantic decoder is then responsible for converting this information back into semantic concepts related to the task.
In an actual power distribution network scene, the measured single-point data packet is very small, and the data packet length is generally kilobits. According to shannon's third theorem, the coding method for adding redundant information can realize higher transmission rate and bring about larger transmission delay. Therefore, the source channel joint coding is needed, the semantic compression ratio, the semantic information merging coding quantity and the data packet length are cooperatively optimized under the normal form of the limited block length, the data transmission efficiency is improved, the error rate is reduced, and the information high-timeliness transmission is realized.
The terminal encodes the information source through semantic extraction and compression. Defining the semantic compression ratio of the terminal d n at the t time slot as x n (t), discretizing the value into K grade, and representing asWherein x n,max and x n,min are respectively the upper and lower limits of the semantic compression ratio, and the semantic compression ratio of the kth gear is expressed asIn order to improve the semantic coding efficiency, a zero-latency strategy is adopted herein, namely, terminal data is subjected to semantic compression coding immediately after being perceived. For simplicity, let d n t be the valid information bit after semantic encoding of each perceived data x n(t)An, where a n is the original data size. In channel coding, a packet length variable is defined as l n (t). The terminal can optimize the semantic information merging coding quantity according to the channel state, the sensing frequency, the sensing data characteristics and the like, namely, the quantity of the semantic information contained in each data packet is adaptively adjusted. Defining the variable of the number of semantic information merging codes as y n(t)＝{1,2,…,j,...yn,max},yn (t) =j means that the time slot d n performs joint channel coding on j pieces of semantic information.
Under the normal form of the source channel joint coding and the finite block length, in order to ensure low transmission delay, the packet length of the $$ t$ time slot needs to meet the following constraint:
ln(t)＞xn(t)yn(t)An (2)
therefore, the transmission delay of the data packet with the length of $l_ { n } (t) $ is:
in the method, in the process of the invention, The channel decoding error probability for the data packet of the tth slot terminal d n,The maximum channel coding rate at the t-th slot for terminal d n is expressed as:
Where γ n,t is the signal-to-noise ratio, R SC(γn,t is d n and the shannon channel capacity of the edge computation gateway, and V (γ n,t) is the channel dispersion, denoted as 1- (1 γ n,t)-1). Is thatIs used as a reference to the remainder of the (c),As a Gaussian functionIs an inverse function of (c).
S1.1.3: information source channel joint decoding model for electric power semantic short packet communication
The edge computing gateway performs channel decoding and semantic decoding on the received compressed semantic coding information. Defining the channel decoding error probability of the $t time slot terminal $d_n$data packet as follows:
The mapping function Ω n,t,j,e =t 'is constructed to represent the jth semantic in the ith data packet decoded by the edge computation gateway in the t-th slot, and the terminal d n performs semantic coding in the t' slot. The semantic decoding error probability is:
where β 1、β2、β3、β4 denotes the model parameters, which can be obtained by Levenberg-Marquardt method learning, which parameters vary with the neural network employed.
S1.2: peak semantic error age for power short packet communication
Definition τ n,t,e,j,Delta n,t,e,j represents the collection time, the semantic decoding completion time and the time interval between the data corresponding to the j-th semantic in the e-th data packet of the terminal d n decoded by the t-th time slot and the next semantic decoding completion time of the edge computing gateway respectively. Definition PAoI n,t,e,j represents the jth semantic PAoI of the edge computing gateway in the ith data packet of the jth slot decoding terminal d n, expressed as:
in the method, in the process of the invention, Representing the end-to-end delay. Thus, PAoI of terminal d n is:
The traditional PAoI indexes are only measured from the angle of information transmission timeliness, and cannot reflect the successful analysis performance of the semantic information of the receiving end. Thus, innovative information timeliness metrics adapted for power short packet communications, PAoIS, are presented herein. PAoIS not only considers the information timeliness of the whole links of data acquisition, semantic coding, short packet transmission and semantic decoding, but also considers the capability of successfully analyzing semantic information of a receiving end and the influence of semantic decoding errors on the state sensing service of the power distribution network. PAoIS is defined as:
In the formula, omega n (t, e, j) represents the importance of the j-th semantic in the e-th data packet of the terminal d n, and reflects the value of the semantic correct analysis on the accurate state sensing service of the power distribution network. Representing decoding semanticsWith true semanticsThere is a deviation between, i.e. a semantic decoding error. σ n represents a semantic decoding error weight, and the larger the value of the semantic decoding error weight is, the greater the semantic decoding error is a main factor causing PAoIS to increase, and a dynamic compromise between semantic information timeliness and semantic information correct decoding can be achieved by adjusting σ n.
The maximum PAoIS obtained by traversing all data packets and semantics is defined as PAoIS of the terminal d n on the edge computing gateway at the t-th time slot, and is expressed as:
constructing epsilon n -PAoIS constraint based on the law of large deviation, ensuring that the probability of the terminal PAoIS exceeding the threshold value is exponentially reduced, which is expressed as:
Wherein PAoIS n,th represents the PAoIS threshold value of terminal d n, ε n is PAoIS the decay rate of the boundary failure probability, and represents the probability that PAoIS of terminal d n exceeds threshold PAoIS n,th as the threshold value increases Descending. Increasing epsilon n can improve the attenuation rate, and provides stricter information timeliness guarantee for accurate state sensing of the power distribution network.
In addition, the second-level accurate state sensing of the power distribution network requires that measurement data of all PMU terminals are on the same information timeliness section, and PAoI deviation constraint needs to be met, which is expressed as:
Where ε' represents the maximum PAoI deviation of the accurate state sensing requirement of the power distribution network.
In a further embodiment, step S2 comprises: optimization problem modeling and optimization problem transformation are two parts. The specific implementation steps are as follows.
S2.1 optimization problem modeling
According to the invention, the high-timeliness and reliable transmission of the multi-mode operation key characteristic information of the power distribution network is realized by jointly optimizing the sensing frequency, the semantic compression ratio, the semantic information merging and encoding quantity and the data packet length and minimizing PAoIS. Defining an optimization variable as f= { f n(i)}、x＝{xn(t)}、y＝{yn(t)}、l＝{ln (t) }, and constructing an optimization problem as follows:
P1:
s.t.C1:
C2:
C3:
C4:
Wherein, C 1 is the constraint of the sensing frequency, the semantic compression ratio and the semantic information merging coding quantity; c 2 is packet length constraint; c 3 is the ε n -PAoIS constraint. C 4 is the PAoI bias constraint.
S2.2 optimization problem transformation
The form of epsilon n -PAoIS constraint in the original optimization problem P1 is extremely complex, and when PAoIS exceeds a threshold PAoIS n,th, the accurate state sensing performance of the power distribution network is significantly affected. Thus, the auxiliary variable H n (t) is introduced to record the condition that PAoIS exceeds the threshold in the t-th time slot, and as an optimization penalty, the update rule is that
In the formula, the indication functionFor monitoring PAoIS n (t) whether the threshold PAoIS n,th is exceeded. The function value is 1 if the threshold is exceeded, and 0 otherwise. exp (- nPAoISn,th) represents a forward penalty increase based on a threshold, decreasing with increasing threshold in an update of H n (t+1), representing a relative increase in the tolerance of the distribution network accurate state awareness to conditions exceeding the threshold. Based on the auxiliary variable H n (t), the optimization objective of P1 can be converted to ψ n (t), expressed as
Where the optimization objective ψ n (t) further considers the additional penalty caused by exceeding the threshold on the basis of PAoIS. Thus, by inter-slot decoupling, the long-term optimization problem P1 is decomposed into a single-slot optimization problem P2, denoted as
P2:
s.t.C1,C2
C4:
Aiming at P2, a 6G sense-of-general integrated resource allocation algorithm suitable for a power semantic short packet transceiver is provided.
S3.1 large-scale perception frequency optimization algorithm driven by electric power in multiple modes
The invention provides a method for capturing multimode operation state changes such as voltage dip, flicker, frequency mutation and the like of a power distribution network on a large scale based on power multimode driving optimization sensing frequency, and realizing accurate sensing of operation situation.
And mining multi-mode operation characteristics of the power distribution network based on PMU measurement data, wherein the Z modes comprise voltage sag, flicker, frequency mutation and the like. The mode indication variable is defined as α n (t), where α n (t) =z represents that the measured data perceived by the terminal d n in the t-th time slot corresponds to the z-th operation mode. Definition of the definitionAndAverage PAoIS, average PAoI and average cost function for terminal d n during period (i-1). The state space of the large-scale perception frequency optimization problem is that
The action space of the terminal d n isConstructing a cost function of d n as
In the running process of the power system, the occurrence probability of various modes is not completely the same, and the occurrence frequency of modes with important influence on the second-level accurate state perception of the power distribution network is extremely low, namely, the situation that part of mode experience samples are sparse exists. The traditional DQN optimization method based on the equal probability sampling or the TD error sampling cannot fully utilize sparse samples, so that the adaptability of a training model and part of sparse modes is poor, the data perception frequency cannot be effectively adjusted and optimized, and the running state change of the power distribution network is difficult to accurately capture. Aiming at the problem, the invention provides an DQN optimization algorithm driven by electric power in a multi-mode manner, a multi-mode experience playback pool is constructed, and an electric power mode exploration-utilization compromise coefficient is introduced, so that a PMU terminal can dynamically adjust the sampling proportion of a multi-mode experience sample according to self learning performance, and the perceived frequency optimization performance under the condition of change of an electric power operation mode is improved.
The concrete introduction is as follows:
(1) DQN network initialization and multi-modal experience playback pool construction
Each terminal initializes its DQN model, including a main networkFor generating data-aware frequency decisions, a target networkFor stabilization training. And respectively constructing Z experience playback sub-pools according to the Z power distribution network modes, wherein the Z experience playback sub-pools are used for storing experience samples of different modes, and the aggregate of the experience samples is denoted as gamma n(i)＝{Υn,1(i),…,Υn,z(i),...,Υn,Z (i).
(2) Perceptual frequency decision and multi-modal experience updating
In the ith large-scale period, the main network is based on an E-greedy strategy and a Q valueSelecting an optimal actionAt the end of the period of time,Transition to the next stateTerminal calculation cost functionFinally, constructing an experience sampleAnd based on alpha n (t) to be stored in the empirical playback sub-pool of the corresponding modality.
(3) Multimode driven DQN network optimization
When the learning performance is poor in a certain electric power mode, the proposed algorithm is biased to be utilized, and the DQN is enhanced to learn the optimal solution in the mode by increasing the extraction probability of the mode experience. On the other hand, the proposed algorithm avoids the optimization performance degradation caused by the poor adaptation of the training model to the sparse mode by encouraging experience in exploring the power modes that occur less frequently. To enhance learning of poor performing modes and increase exploration of sample sparse modes, power mode exploration is defined-with trade-off coefficients κ n (i) for adjusting sampling strategies, larger for the larger representative of κ n being more prone to exploration. In the ith large scale period, the empirical importance weight φ n,z (i) of the z-th modal empirical playback sub-pool of terminal d n is represented as
Where TD n,z,o represents the TD error for the o-th experience, |y n,z (i) | represents the number of experiences in the z-th experience playback sub-pool, and ||y n (i) | represents the total number of experiences in all experience playback pools.
The playback experience sample set phi n (i) is extracted by a weighted sampling method and based on importance weights, and based on this, network training is performed. The loss function xi n (i) is calculated as
In the method, in the process of the invention,Terminal d n representing target network evaluation in stateThe minimum value of the Q values corresponding to all the next possible actions f n (i'),Is a discount factor.
Finally, the terminal d n optimizes the main network based on the gradient descent method and the loss function xi n (i)Every I 0 large time scales, updating the target network to be
S3.2 small-scale information source channel joint coding robust optimization algorithm based on cooperative game DAC
The small scale source channel joint coding optimization decisions of different PMU terminals are mutually coupled. On the one hand, since multiple terminals share computing resources of the edge computing gateway, allocating excessive computing resources for a certain terminal for semantic information decoding can reduce PAoI, but can lead to PAoI rise of other terminals. On the other hand, PAoI deviation constraint defined by formula (12) requires collaborative optimization between multiple terminals, thereby ensuring that the measurement data of each terminal is located on the same PAoI section. In addition, the invention considers the uncertainty caused by channel fluctuation, each terminal needs to ensure the robustness of the channel coding optimization of the information source, and PAoI deviation constraint is satisfied while PAoIS is reduced. The traditional multi-agent deep reinforcement learning algorithm ignores disturbance of channel uncertainty on learning performance and cooperative game relation among multiple terminals, so that multi-agent cooperative optimization is difficult to realize, and joint coding robustness is insufficient.
Aiming at the challenges, the invention provides a small-scale source channel joint coding robust optimization method based on a cooperative game DAC. Firstly, a paradigm of fusion of centralized offline learning and distributed online decision is constructed, on one hand, the terminal makes online decision by using an offline training model, and the joint coding performance deterioration caused by cold start can be avoided. On the other hand, the edge computing gateway builds an experience pool based on different terminal distributed online decisions, so that multi-terminal optimal decision knowledge sharing is realized, and offline training performance is improved. And secondly, disturbance caused by uncertainty of an anti-intelligent simulation channel to learning is introduced, and each terminal learns reliable decisions in an uncertain scene through interaction with the anti-intelligent agent, so that the robustness of joint coding is ensured. Finally, by jointly considering PAoIS performance and PAoI deviation of each terminal, an countermeasure-cooperation cost function is constructed for offline training action networks and joint evaluation networks, so that cooperation game relations among different terminal decisions are effectively mined, cooperation among multiple agents is realized, robustness of joint coding of an information source channel is guaranteed, and PAoI deviation constraint is guaranteed while each terminal PAoIS is minimized under an uncertain scene.
(1) Anti-partnership POMDP construction
The problem of joint coding optimization of small-scale source channels is built into an anti-cooperation POMDP model, and the tuple is defined as
In the method, in the process of the invention,And (3) withRepresenting the state space, the set of action spaces, respectively, P ini is the conditional transition probability between the initial states,AndRepresenting an uncertainty set of all state transition probabilities and cost functions.AndRespectively a partially observable state and a conditional probability of that observable state. η is a discount factor.
(2) Algorithm execution flow
The proposed algorithm comprises an offline centralized learning phase and an online distributed learning phase:
1) Offline centralized learning phase
Step one: defining an anti-agent, an offline action network, and a joint evaluation network asAndThe edge computing gateway extracts H groups of experiences from the experience pool for offline training. Wherein the decision of the antibody intelligent network and the action network based on the experience of the terminal d n in the h group is expressed as
In the method, in the process of the invention,Representing the observed state space of terminal d n in the h-th set of experience, Δγ n,h represents the actions of the anti-agent on the anti-agent network during the offline centralized learning phase.
Step two: in the h sample, the combined consideration PAoI and PAoIS builds an action network with a challenge-co cost function of
Step three: constructing TD error of action network, countermeasure network and joint evaluation network as
In the method, in the process of the invention,Indicating the system state after the action is performed.
Step four: based on TD error, the gradient descent method is utilized to train and update the action networkAnd an anti-agent networkRepresented as
Joint evaluation networkTraining updates are expressed as
2) Online distributed learning phase
Step one: each terminal uses the off-line training action network issued by the edge computing gateway as a local on-line action network, namelyAnd based on this, make online optimization decisions, denoted as
Step two: observing state change of each terminal, and constructing an experience sampleAnd put into an experience pool. Wherein,Indicating that terminal d n is policy basedAnd performing the optimal action of joint coding of the source channels.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, and it should be noted that it is possible for those skilled in the art to make several improvements and modifications without departing from the technical principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.
Claims (13)
1. A power semantic short packet communication method based on general sense integration is characterized in that: the method comprises the following specific steps:
Constructing a power semantic short packet communication system model based on communication integration, wherein the power semantic short packet communication system model comprises a PMU measurement data sensing model, a source channel joint coding model, a source channel joint decoding model and a peak semantic error age model facing power short packet communication;
Constructing a multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem oriented to power semantic short packet communication based on the power semantic short packet communication system model, and converting the multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem oriented to power semantic short packet communication into a single-time-slot optimization problem through decoupling among time slots;
The large-scale sensing frequency optimization algorithm driven by the electric power in a multi-mode and the small-scale information source channel joint coding robust optimization algorithm based on the cooperative game DAC are provided, a single-time slot optimization problem is solved, and the information source channel joint coding of accurate sensing of the running situation and electric power semantic short packet communication is realized.
2. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the PMU measurement data perception model specifically comprises:
Defining the measured data sensing frequency of the PMU terminal d n in the period i as f n (i), discretizing the value of f n (i) into M-grade, and representing as:
wherein F n,max and F n,min are respectively the upper and lower limits of the sensing frequency, and the m-th gear sensing frequency is The total optimized time length is divided into I large time scales, namely time periods, and the aggregate isEach period is composed of T 0 small time scale time slots, wherein the time slot set of the ith period isThe total number of slots is t=it 0; define the number of PMU terminals as N and aggregate as
3. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the source channel joint coding model specifically comprises the following steps:
Defining the semantic compression ratio of the terminal d n at the t time slot as x n (t), discretizing the value into K grade, and representing as Wherein x n,max and x n,min are respectively the upper and lower limits of the semantic compression ratio, and the semantic compression ratio of the kth gear is expressed as
Defining the valid information bit of the data perceived by d n each time after semantic coding as x n(t)An, wherein A n is the original data size; in channel coding, defining a packet length variable as l n (t); defining the variable of the combined coding quantity of the semantic information as y n(t)＝{1,2,…,j,…yn,max},yn (t) =j to represent that the time slot d n carries out joint channel coding on j semantic information; under the normal form of the joint coding of the information source channel and the finite block length, in order to ensure low transmission delay, the data packet length of t time slots needs to meet the following constraint:
ln(t)＞xn(t)yn(t)An (2)
Therefore, the transmission delay of the data packet with the length l_ { n } (t) is:
in the method, in the process of the invention, The channel decoding error probability for the data packet of the tth slot terminal d n,The maximum channel coding rate at the t-th slot for terminal d n is expressed as:
Where γ n,t is the signal-to-noise ratio, R SC(γn,t) is d n and the shannon channel capacity of the edge computation gateway, V (γ n,t) is the channel dispersion, expressed as 1- (1 γ n,t)-1; Is that Is used as a reference to the remainder of the (c),As a Gaussian functionIs an inverse function of (c).
4. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the source channel joint decoding model specifically comprises the following steps:
defining the channel decoding error probability of the data packet of the terminal d_n of the t time slot as follows:
Constructing a mapping relation function omega n,t,j,e = t ', representing the j-th semantic in the e-th data packet decoded by the edge computing gateway in the t-th time slot, and carrying out semantic coding by the terminal d n in the t' time slot; the semantic decoding error probability is:
where β 1、β2、β3、β4 denotes the model parameters, which can be learned by the Levenberg-Marquardt method, and which vary with the neural network employed.
5. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the peak semantic error age model for power short packet communication specifically comprises the following steps:
definition τ n,t,e,j, Delta n,t,e,j represents the collection time, the semantic decoding completion time and the time interval between the data corresponding to the j-th semantic in the e-th data packet of the terminal d n decoded by the t-th time slot and the next semantic decoding completion time of the edge computing gateway respectively; definition PAoI n,t,e,j represents the jth semantic PAoI of the edge computing gateway in the ith data packet of the jth slot decoding terminal d n, expressed as:
in the method, in the process of the invention, Representing end-to-end delay; thus, PAoI of terminal d n is:
defining an innovative information timeliness measurement index suitable for power short packet communication, namely PAoIS is as follows:
wherein ω n (t, e, j) represents the importance of the j-th semantic in the e-th packet of the terminal d n; Representing decoding semantics With true semanticsDeviation exists between the two, namely semantic decoding errors; σ n denotes semantic decoding error weights;
the maximum PAoIS obtained by traversing all data packets and semantics is defined as PAoIS of the terminal d n on the edge computing gateway at the t-th time slot, and is expressed as:
constructing epsilon n -PAoIS constraint based on the law of large deviation, ensuring that the probability of the terminal PAoIS exceeding the threshold value is exponentially reduced, which is expressed as:
Wherein PAoIS n,th represents the PAoIS threshold value of terminal d n, ε n is PAoIS the decay rate of the boundary failure probability, and represents the probability that PAoIS of terminal d n exceeds threshold PAoIS n,th as the threshold value increases Descending;
in addition, the second-level accurate state sensing of the power distribution network requires that measurement data of all PMU terminals are on the same information timeliness section, and PAoI deviation constraint is met, which is expressed as:
Where ε' represents the maximum PAoI deviation of the accurate state sensing requirement of the power distribution network.
6. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the multi-time scale heterogeneous resource high-timeliness collaborative optimization problem modeling for the power semantic short packet communication is as follows:
Wherein, C 1 is the constraint of the sensing frequency, the semantic compression ratio and the semantic information merging coding quantity; c 2 is packet length constraint; c 3 is the ε n -PAoIS constraint; c 4 is PAoI bias constraint; the optimization variable is defined as f= { f n(i)}、x＝{xn(t)}、y＝{yn(t)}、l＝{ln (t) }.
7. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the single time slot optimization problem is specifically as follows:
Introducing auxiliary variable H n (t) to record the condition that PAoIS exceeds a threshold value in the t time slot, and taking the condition as an optimization penalty, wherein the update rule is as follows
In the formula, the indication functionFor monitoring PAoIS n (t) if the threshold PAoIS n,th is exceeded; if the threshold value is exceeded, the function value is 1, otherwise, the function value is 0; exp (- ε nPAoISn,th) represents a forward penalty increment based on a threshold, decreases with increasing threshold in the update of H n (t+1), and represents a relative increase in the tolerance of the accurate state sensing of the power distribution network to conditions exceeding the threshold; based on the auxiliary variable H n (t), the optimization objective of P1 can be converted to ψ n (t), expressed as
Wherein the optimization objective ψ n (t) further considers the additional penalty caused by exceeding the threshold on the basis of PAoIS; thus, by inter-slot decoupling, the long-term optimization problem P1 is decomposed into a single-slot optimization problem P2, denoted as
8. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the large-scale perception frequency optimization algorithm for the electric power multi-mode driving specifically comprises the following steps:
Based on PMU measurement data, mining multi-mode operation characteristics of the power distribution network, including Z modes such as voltage sag, flicker, frequency mutation and the like; defining a mode indicating variable as alpha n (t), wherein alpha n (t) =z represents that the measured data perceived by the terminal d n in the t time slot corresponds to the z-th operation mode; definition of the definition AndAverage PAoIS, average PAoI and average cost function for terminal d n during period (i-1); the state space of the large-scale perception frequency optimization problem is that
The action space of the terminal d n isConstructing a cost function of d n as
(1) DQN network initialization and multi-modal experience playback pool construction
Each terminal initializes its DQN model, including a main networkFor generating data-aware frequency decisions, a target networkFor stabilization training; constructing Z experience playback sub-pools respectively according to Z power distribution network modes, wherein the Z experience playback sub-pools are used for storing experience samples of different modes, and the aggregate of the experience samples is denoted as gamma n(i)＝{Υn,1(i),…,Υn,z(i),…,Υn,Z (i);
(2) Perceptual frequency decision and multi-modal experience updating
In the ith large-scale period, the main network is based on an E-greedy strategy and a Q valueSelecting an optimal actionAt the end of the period of time,Transition to the next stateTerminal calculation cost functionFinally, constructing an experience sampleAnd based on alpha n (t), storing the data into an experience playback sub-pool of the corresponding mode;
(3) Multimode driven DQN network optimization
Defining a power modality exploration-utilizing a trade-off factor κ n (i) for adjusting sampling strategies, the larger κ n represents the more exploration prone; in the ith large scale period, the empirical importance weight φ n,z (i) of the z-th modal empirical playback sub-pool of terminal d n is represented as
Where TD n,z,o represents TD error of the o-th experience, |y n,z (i) | represents the number of experiences in the z-th experience playback sub-pool, |y n (i) | represents the total number of experiences in all experience playback pools;
Extracting a playback experience sample set phi n (i) by using a weighted sampling method and based on importance weight, and performing network training based on the playback experience sample set phi n (i); the loss function xi n (i) is calculated as
In the method, in the process of the invention,Terminal d n representing target network evaluation in stateThe minimum value of the Q values corresponding to all the next possible actions f n (i'),Is a discount factor;
Finally, the terminal d n optimizes the main network based on the gradient descent method and the loss function xi n (i) Every I 0 large time scales, updating the target network to be
9. The power semantic short packet communication method based on the sense integration according to claim 1, wherein the method is characterized in that: the small-scale information source channel joint coding robust optimization algorithm based on the cooperative game DAC specifically comprises the following steps:
(1) Anti-partnership POMDP construction
The problem of joint coding optimization of small-scale source channels is built into an anti-cooperation POMDP model, and the tuple is defined as
In the method, in the process of the invention,And (3) withRepresenting the state space, the set of action spaces, respectively, P ini is the conditional transition probability between the initial states,AndAn uncertainty set representing all state transition probabilities and cost functions; And Respectively, a part of observable states and conditional probabilities of the observable states; η is a discount factor;
(2) Algorithm execution flow
The proposed algorithm comprises an offline centralized learning phase and an online distributed learning phase:
1) Offline centralized learning phase
Step one: defining an anti-agent, an offline action network, and a joint evaluation network asAndThe edge computing gateway extracts H groups of experiences from the experience pool to perform offline training; wherein the decision of the antibody intelligent network and the action network based on the experience of the terminal d n in the h group is expressed as
In the method, in the process of the invention,Representing the observation state space of the terminal d n in the h group of experience, Δγ n,h representing the actions of the anti-agent on the anti-agent network in the offline centralized learning phase;
Step two: in the h sample, the combined consideration PAoI and PAoIS builds an action network with a challenge-co cost function of
Step three: TD errors of constructing action network, countermeasure network and joint evaluation network are as follows
In the method, in the process of the invention,Representing a system state after the action is performed;
Step four: based on TD error, training and updating action network by gradient descent method And an anti-agent networkRepresented as
Joint evaluation networkTraining updates are expressed as
2) Online distributed learning phase
Step one: each terminal uses the off-line training action network issued by the edge computing gateway as a local on-line action network, namelyAnd based on this, make online optimization decisions, denoted as
Step two: observing state change of each terminal, and constructing an experience sampleAnd placing the model into an experience pool; wherein,Indicating that terminal d n is policy basedAnd performing the optimal action of joint coding of the source channels.
10. Electric power semantic short packet communication system based on sense integration, characterized in that: the device comprises a terminal layer and an edge layer;
The terminal layer comprises a power semantic short packet communication PMU terminal, and the power semantic short packet communication PMU terminal comprises:
And a data acquisition module: the method comprises the steps of collecting operation state data of electrical equipment;
Semantic coding module: the method comprises the steps of preprocessing and encoding collected original data based on service characteristics, and extracting key information, namely semantics, in the data;
the semantic compression module: the method is used for compressing the encoded semantic information so as to reduce the bandwidth required by data transmission;
And the information source channel joint coding module: the method is used for combining source coding, namely semantic coding and compression, and channel coding, adding redundant information to resist noise and interference in a channel, and improving the reliability of data transmission;
And a communication module: the method comprises the steps of sending information such as semantic compression ratio, semantic information merging coding quantity, data packet length and the like to a power semantic short packet communication edge computing gateway, and receiving information such as maximum PAoIS, average PAoIS, average PAoI, PAoIS boundary damage probability attenuation rate and the like issued by the power semantic short packet communication edge computing gateway;
And the general sense integrated calculation module is as follows: the system is used for analyzing and processing the acquired data in real time, supporting the integrated optimization of sensing and communication and dynamically adjusting the acquisition frequency;
and a power supply module: for providing stable power support for the terminal device;
The edge layer comprises a 6G base station and an electric power semantic short packet communication edge computing gateway; the power semantic short packet communication edge computing gateway comprises:
and a channel decoding module: the PMU terminal is used for processing the data packet received from the PMU terminal, carrying out signal detection and channel decoding, recovering compressed semantic information and ensuring the integrity and reliability of data;
semantic decoding module: the system comprises a channel decoding module, a power system operation state generation module, a compression semantic information generation module and a power system operation state generation module, wherein the channel decoding module is used for receiving compressed semantic information output by the channel decoding module and converting the compressed semantic information back to original semantics related to the power system operation state; the step involves analyzing semantic codes, reconstructing key information of original data, and providing a basis for subsequent data processing and analysis;
And a communication module: the PMU terminal is used for exchanging data with the PMU terminal; supporting high-speed data receiving and sending through a 6G network, sending information such as maximum PAoIS and average PAoIS, average PAoI, PAoIS boundary failure probability attenuation rate and the like to a power semantic short packet communication PMU terminal, and receiving information such as semantic compression ratio, semantic information merging coding quantity, data packet length and the like sent by the power semantic short packet communication PMU terminal;
Peak semantic error age calculation module: PAoIS for calculating each data packet based on the received data, and evaluating the timeliness and accuracy of the data; the module is important to ensure that the data processing meets the accurate state sensing requirement of the power distribution network;
And a data storage module: the power distribution network key circuit transient state data information acquisition module is used for storing power semantic short packet communication PMU terminal acquired power distribution network key circuit transient state data information and providing support for state analysis and fault diagnosis;
And a data processing module: the method is used for further analyzing and processing the decoded data and providing data support for second-level accurate state sensing and operation trend early warning and fault diagnosis positioning of the power distribution network;
And a power supply module: for providing a stable and continuous power supply for a power semantic short packet communication edge computing gateway, ensuring that the system can operate uninterrupted.
11. Electric power semantic short packet communication device based on sense integration, its characterized in that: comprising
Modeling module: the method is used for constructing a power semantic short packet communication system model based on communication sense integration, and comprises a PMU measurement data sensing model, a source channel joint coding model, a source channel joint decoding model and a peak semantic error age model facing power short packet communication;
And a conversion module: the method is used for constructing a multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem for power semantic short packet communication based on the power semantic short packet communication system model, and converting the multi-time-scale heterogeneous resource high-timeliness collaborative optimization problem for power semantic short packet communication into a single-time-slot optimization problem through time-slot decoupling;
And a solving module: the method is used for solving a single-time-slot optimization problem by providing a large-scale perception frequency optimization algorithm driven by electric power multi-mode and a small-scale information source channel joint coding robust optimization algorithm based on a cooperative game DAC, and realizing information source channel joint coding of accurate perception of running situation and electric power semantic short packet communication.
12. The power semantic short packet communication device based on sense of general integration according to claim 11, wherein: in the conversion module, the multi-time scale heterogeneous resource high timeliness collaborative optimization problem facing the power semantic short packet communication is modeled as follows:
Wherein, C 1 is the constraint of the sensing frequency, the semantic compression ratio and the semantic information merging coding quantity; c 2 is packet length constraint; c 3 is the ε n -PAoIS constraint; c 4 is PAoI bias constraint; the optimization variable is defined as f= { f n(i)}、x＝{xn(t)}、y＝{yn(t)}、l＝{ln (t) }.
13. The power semantic short packet communication device based on sense of general integration according to claim 11, wherein: in the conversion module, the single-time slot optimization problem is specifically as follows:
Introducing auxiliary variable H n (t) to record the condition that PAoIS exceeds a threshold value in the t time slot, and taking the condition as an optimization penalty, wherein the update rule is as follows
In the formula, the indication functionFor monitoring PAoIS n (t) if the threshold PAoIS n,th is exceeded; if the threshold value is exceeded, the function value is 1, otherwise, the function value is 0; exp (- ε nPAoISn,th) represents a forward penalty increment based on a threshold, decreases with increasing threshold in the update of H n (t+1), and represents a relative increase in the tolerance of the accurate state sensing of the power distribution network to conditions exceeding the threshold; based on the auxiliary variable H n (t), the optimization objective of P1 can be converted to ψ n (t), expressed as
Ψn(t)＝PAoISn(t)+Hn(t)M{PAoISn(t)＞PAoISn,th} (15)
Wherein the optimization objective ψ n (t) further considers the additional penalty caused by exceeding the threshold on the basis of PAoIS; thus, by inter-slot decoupling, the long-term optimization problem P1 is decomposed into a single-slot optimization problem P2, denoted as
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118338321A true CN118338321A (en) | 2024-07-12 |
Family
ID=
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111062464B (en) | Power communication network reliability prediction and guarantee method and system based on deep learning | |
| KR102439722B1 (en) | Weather data-based wireless sensor network node solar energy collection power prediction algorithm | |
| CN102316496A (en) | Data merging method based on Kalman filtering in wireless sensor network | |
| CN111710150A (en) | Abnormal electricity consumption data detection method based on countermeasure self-coding network | |
| Luo et al. | Link quality estimation method for wireless sensor networks based on stacked autoencoder | |
| CN103763706A (en) | Collaborative spectrum sensing vague fusion method based on clustering | |
| CN114158102B (en) | Wireless heterogeneous communication network switching method for feeder automation real-time control | |
| CN115469627B (en) | Intelligent factory operation management system based on Internet of things | |
| CN116415705A (en) | Multi-factor water level prediction method and device, electronic equipment and computer medium | |
| CN103532673B (en) | Distributed wireless meteorological encoding monitoring method, device and system | |
| CN111882119A (en) | Battery SOH prediction optimization method based on SA-BP neural network | |
| CN117459188B (en) | Electric Beidou communication system and communication method based on Beidou communication technology | |
| Wang | Application of Wireless Sensor Network based on LoRa in City Gas Meter Reading. | |
| CN107437976A (en) | A kind of data processing method and equipment | |
| CN118338321A (en) | Electric power semantic short packet communication method, device and system based on general sense integration | |
| Yang et al. | Vehicle text data compression and transmission method based on maximum entropy neural network and optimized huffman encoding algorithms | |
| CN112118038A (en) | Satellite efficiency evaluation method | |
| CN116961674A (en) | Distributed photovoltaic data compression method and system | |
| CN114124725A (en) | Quantum communication network reliability comprehensive evaluation method based on complex network model | |
| CN111950928B (en) | Loss reduction method and device for power distribution network, storage medium and computing equipment | |
| CN115048982A (en) | Intelligent power grid monitoring system for data center and control method thereof | |
| CN112782584B (en) | Method, system, medium and device for predicting remaining usage amount of battery electric quantity | |
| CN108667757A (en) | Manchester code sampling processing method, module and decoder | |
| CN115496175A (en) | Newly-built edge node access evaluation method and device, terminal equipment and product | |
| CN108449775B (en) | Method for selecting relay node of radio frequency energy capture network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication |