Edge container-based electric carbon computing intelligent fusion terminal and time synchronization method

Abstract

This invention provides an intelligent fusion terminal for carbon computing based on edge containers and a time synchronization method, comprising a physical layer, a protocol interface layer, and an application layer; including the following steps: (1) system model construction; (2) optimization problem modeling; (3) low-cost and accurate topology perception optimization for smart parks based on redundancy-removed and accurate BP neural networks; (4) efficient DQN time synchronization and timing data packet transmission optimization based on Q-difference; (5) optimization of carbon computing task offloading and resource allocation based on exchange matching. This invention uses a redundancy-removed and accurate BP neural network to complete topology prediction in the topology perception container of the intelligent fusion terminal, and proposes an efficient DQN time synchronization and timing data packet transmission optimization based on Q-difference, improving the time synchronization accuracy of smart parks; through optimization of carbon computing task offloading and resource allocation based on exchange matching, it achieves efficient transmission and calculation of carbon information of smart park equipment, supporting the efficient and stable operation of low-carbon services.

Description

Technical Field

[0001] This invention belongs to the field of information technology, specifically relating to an edge container-based intelligent fusion terminal for carbon computing and a time synchronization method. Background Technology

[0002] A smart industrial park refers to an industrial park characterized by digitalization, intelligence, and energy conservation, achieved through next-generation information and communication technologies and digital technologies. By deploying numerous renewable energy devices such as photovoltaic power plants and wind turbines, smart parks optimize load and energy regulation strategies, promote energy conservation and emission reduction in the power system, and achieve low-carbon operation. Low-carbon operations in smart parks, such as carbon metering, carbon footprint monitoring, and carbon trading, require the acquisition, screening, integration, processing, and analysis of electricity carbon information to understand the operational status of the smart park and adjust electricity carbon emission strategies. Therefore, electricity carbon calculation is a crucial foundation for the accurate, stable, and reliable operation of low-carbon operations in smart parks. To ensure the accuracy of electricity carbon calculation, on the one hand, it is necessary to ensure the time consistency of electricity carbon information collection, accurately reflecting the real-time status of electricity carbon emissions in the smart park; on the other hand, low-latency electricity carbon calculation technology is needed to reduce the transmission and calculation latency of electricity carbon information.

[0003] Time synchronization, a crucial technological support for smart parks, provides accurate time information for carbon dioxide calculations. Many business terminal devices cannot directly connect to 5G time synchronization networks, State Grid time synchronization networks, or BeiDou satellites; they only connect to intelligent fusion terminals via power line carrier signals, making accurate time synchronization difficult. Intelligent fusion terminals need to integrate multiple clocks, including BeiDou satellites and 5G time synchronization networks, to obtain time synchronization information, and also need to act as a new clock source to provide time synchronization for numerous devices in the smart park. Edge computing, by pushing computing resources closer to the network edge where business operations are conducted, can significantly reduce communication distance and data transmission latency by connecting edge intelligent terminals to servers; furthermore, it can reduce data processing latency by utilizing the computing resources on servers. By integrating time synchronization and edge computing, deploying terminals with clock sources at the network edge can solve the problem of large timing errors caused by long-term transmission of time synchronization data packets, improving the time consistency of carbon dioxide information collection.

[0004] Building upon time synchronization and edge computing, accurate calculation of carbon dioxide information also requires topology sensing and optimization of carbon dioxide calculation offloading strategies. However, currently, the various functions required for carbon dioxide calculation are typically implemented by different terminals, necessitating collaboration between them. This results in significant information exchange, leading to increased latency and decreased accuracy in carbon dioxide calculation. Therefore, there is an urgent need to design a smart park intelligent fusion terminal that integrates topology sensing, time synchronization, offloading strategy optimization, and carbon dioxide calculation functions. This would improve park time synchronization accuracy while reducing carbon dioxide calculation latency and further enhancing the accuracy of carbon dioxide calculation. However, existing fusion terminals and time synchronization methods still face the following challenges:

[0005] First, integrating terminal topology awareness presents challenges. Topology awareness is fundamental for time synchronization and carbon dioxide computation. Smart parks utilize carrier communication, making the network topology susceptible to external environmental factors and impedance fluctuations. Traditional topology awareness methods, typically employing path search, are ill-suited to the variable topology of smart parks, leading to prolonged path search times, inaccurate topology information analysis, and ultimately, reduced reliability of park time synchronization and increased latency in carbon dioxide information transmission.

[0006] Secondly, the time synchronization speed is slow and the accuracy is low. The complex network environment and variable topology of the campus make it difficult for existing sensing device timing strategies to identify the topology and select the optimal transmission path for timing data packets, resulting in slow time synchronization. Furthermore, because carrier communication is susceptible to electromagnetic interference and channel conditions change rapidly, it is difficult to accurately obtain the time overhead for the fusion terminal to synchronize with the sensing devices, leading to low time synchronization accuracy. Existing timing and synchronization strategies are not applicable to the smart fusion terminal for carbon computing.

[0007] Third, the task offloading strategy for carbon dioxide information is coupled with the server computing resource allocation strategy. Because carbon dioxide information task offloading requires minimal offloading latency, when formulating the task offloading strategy, sensing devices tend to select servers with closer proximity, better channel conditions, and more available computing resources. Therefore, servers with more computing resources are more likely to receive task offloading requests. However, servers allocate computing resources based on the task offloading requests from sensing devices. When too many tasks are offloaded to a single server, the computing latency increases significantly, leading to increased carbon dioxide computation latency. Therefore, optimizing the task offloading strategy while ensuring transmission latency, reducing computation latency, and resolving the coupling problem between the carbon dioxide information task offloading strategy and the server computing resource allocation strategy is a challenge. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent fusion terminal for carbon calculation based on edge containers and a time-synchronized carbon calculation method thereof. Through this fusion terminal or time-synchronized carbon calculation method, efficient and accurate distributed calculation of carbon can be achieved, with high time synchronization accuracy, low calculation latency, and high calculation precision.

[0009] According to specific embodiments of the present invention, the present invention can further solve the following technical problems:

[0010] In existing carbon computing, inaccurate topology awareness leads to a lack of accurate topology information in timing data packet transmission and carbon information task offloading strategies.

[0011] Existing carbon-based computing suffers from problems such as smart park topology changes and electromagnetic interference significantly impacting timing errors.

[0012] Existing carbon computing suffers from the problem of task offloading being coupled with server resource allocation, making it difficult to formulate an excellent task offloading strategy.

[0013] The present invention first provides the following technical solution:

[0014] A method for calculating carbon emissions based on a smart fusion terminal with edge containers, comprising:

[0015] S1 obtains the system's topology information, wherein the topology information includes channel state information and a topology connection matrix containing the connection relationships between the intelligent fusion terminal, sensing devices, and edge servers;

[0016] S2 synchronizes the timing of each sensing device according to the topology information by optimizing the operation method, so that each sensing device has time consistency.

[0017] After achieving time consistency, S3 obtains the carbon information task offloading strategy and the carbon computing resource allocation scheme. The sensing device offloads the computing information according to the carbon information task offloading strategy, and the edge server completes the allocated computing resources according to the carbon computing resource allocation scheme to obtain low-latency carbon computing results.

[0018] The optimized operation method is obtained by solving the following optimization problems: (1) minimizing the average time synchronization cumulative error of each iteration; (2) minimizing the total delay of the electric carbon calculation.

[0019] According to some preferred embodiments of the present invention, the topology information is obtained by a two-layer neural network composed of a SOM network and a BP network. The two-layer neural network includes: a first-layer neural network formed by the SOM network for effective data classification and filtering, and a second-layer neural network formed by the BP network for topology awareness based on the effective data obtained from the first-layer neural network; wherein the SOM network includes an input layer and a competition layer (i.e., its output layer), and the BP network includes a hidden layer and an output layer; the process of obtaining the topology information through this two-layer neural network includes:

[0020] The first layer of the neural network is trained based on a historical electrical database data sample group containing labels for normal data sample groups and abnormal data sample groups.

[0021] All electrical data samples from the sensing devices received in the current time slot are input into the first layer of the trained neural network for classification, and data classified as abnormal sample groups are removed.

[0022] The data classified as normal data sample group is input into the second layer neural network to perceive the topological connection relationship under the current time slot, and the accuracy of the topological perception of the second layer neural network is continuously adjusted and improved until a high-accuracy second layer neural network is obtained.

[0023] The obtained high-precision second-layer neural network outputs topological information based on the data from the normal data sample group.

[0024] According to some preferred embodiments of the present invention, the classification includes:

[0025] (1) In the current iteration of the SOM neural network, the winning neuron in its competitive layer is determined by minimizing the Euclidean distance between itself and the neurons in its input layer, wherein the Euclidean distance is calculated using the following formula:

[0026]

[0027] Among them, w i,j (e) represents the connection weight between input layer neuron i and competing layer neuron j in the current e-th iteration; y i (e) represents electrical data sample group y n The input vector Y = (y1(e),...,y2(e),...,y) = <ν,θ> is (t) = <ν,θ>. n (e)) T In the i-th element, ν represents the sample group of equipment operating voltage phase angle data, θ represents the sample group of equipment operating injected power data, and e represents the set of iterations when the SOM neural network performs classification: E = (1,...,e,...,E). TIn the context of the e-th iteration, E represents the total number of iterations, and T represents the transpose;

[0028] (2) After obtaining the winning neuron, the connection weights between the input layer neuron i and the competing layer neuron j are continuously adjusted during the iteration, as follows:

[0029] w i,j (e+1)=w i,j (e)+χ(e)h(e)(y i,t -w i,j (e)) (14)

[0030] Where 0 < χ(e) < 1 is the learning rate function; h(e) is the neighborhood function related to the data radius, and its value decays over time;

[0031] Repeat the iterative process until the neural network learning rate drops to 0, obtaining the data radius of the electrical data sample group of all sensing devices, and then removing those exceeding the data radius constraint r. min Data sample groups were marked as outliers and removed.

[0032] According to some preferred embodiments of the present invention, obtaining the output topology information includes:

[0033] The normal electrical data sample group after being classified and eliminated by the SOM neural network is input into the hidden layer of the BP neural network;

[0034] By continuously updating the network relationship using the actual input and desired output of the BP neural network, the perceived channel state information is obtained. and topology connection matrix ,in, This includes the connection relationships between intelligent converged terminals and devices, as well as the connection relationships between edge servers and devices, as follows:

[0035]

[0036] Where N0 represents channel noise power information, ζ represents channel electromagnetic interference information, and Z... n→n',t Z is an indicator variable for the topology connections between sensing devices. n→m,t For sensing devices and edge servers m Inter-topology connection indicator variable.

[0037] According to some preferred embodiments of the present invention, the synchronization time is accomplished by a DQN network based on an efficient DQN algorithm using Q-difference, specifically including:

[0038] Based on the empirical data derived from the Q-difference, empirical data is preferentially extracted from the empirical pool according to the probability.

[0039] Based on the extracted empirical data, deep reinforcement learning of neural networks based on the DQN architecture is carried out through the DQN iterative algorithm.

[0040] The timing data packet transmission strategy is output by the neural network based on the DQN architecture. The timing data packet transmission strategy includes selecting timed devices with existing topology links and small cumulative timing errors for untimed devices to send timing data packets to them.

[0041] Each sensing device completes device time synchronization according to the time synchronization data packet transmission strategy;

[0042] The deep reinforcement learning includes: obtaining the state in each iteration through a neural network based on the DQN architecture. Next, take action n',t The Q value, i.e., the estimated value of Q. Subsequently, by continuously updating the neural network parameters, the estimated value of the obtained Q-value is improved. By approximating the actual value, a neural network that can achieve precise action output under state input is finally obtained;

[0043] Among them, the empirical data priority extraction probability based on Q difference prob k The settings are as follows:

[0044]

[0045] Among them, Q k For the estimated Q value of the k-th set of empirical data, Let Q be the target network Q-value.

[0046] According to some preferred embodiments of the present invention, the state space, action space, and reward function in the DQN iterative algorithm are respectively set as follows:

[0047] State space:

[0048]

[0049] in, Represents the topology connection matrix. This indicates the size of the timing data packet obtained in each iteration;

[0050] Action space:

[0051]

[0052] Among them, act n′，t This represents the set of timing data packet transmission strategies formulated by the intelligent fusion terminal for the sensing device, u n Indicates that the time synchronization device, u n'Indicates a device that has not received a time synchronization signal; Z represents a collection of intelligent converged terminals and devices; n→n',t Indicates the topology connection indicator variable between devices; x n→n',t For the t-th iteration, the intelligent fusion terminal or the time-synchronized device u n n = 0, 1, ..., N pairs of unsynchronized devices u n' The timing completion value for n' = 1, 2, ..., N is specifically: when the t-th iteration occurs, the non-timing device u n' Obtain intelligent converged terminal or time-synchronized device u n If it is granted, then x n→n',t =1, otherwise x n→n',t =0;

[0053] Reward function:

[0054]

[0055] in, For the untimed device u in the t-th iteration n' From the time-synchronized device u n The cumulative synchronization error of the received time synchronization data packets satisfies:

[0056]

[0057] The first term is u n The cumulative synchronization error, the second term is u n Transmit time synchronization data packets to u n' The resulting single-slot error.

[0058] According to some preferred embodiments of the present invention, the optimized operation method is obtained by:

[0059] (1) Construct an electric carbon computing system model containing the intelligent fusion terminal, the edge server, the sensing device and their interconnections;

[0060] (2) Construct a time synchronization data packet transmission model based on the synchronous time synchronization strategy under the electric carbon computing system model;

[0061] (3) Based on the time synchronization data packet transmission model, construct a time synchronization error model under the electric carbon computing system model;

[0062] (4) Based on the time-synchronization data packet transmission model, construct the electric carbon information transmission model under the electric carbon calculation system model;

[0063] (5) Based on the time synchronization error model and the electric carbon information transmission model, construct the total delay model of electric carbon calculation of the electric carbon calculation system model;

[0064] (6) Based on the electric carbon computing system model, the time synchronization data packet transmission model, the time synchronization error model, the carbon information transmission model, and the electric carbon computing total delay model, the optimization problem is set, and the optimization problem is solved by the electric carbon information task unloading and computing resource allocation calculation based on the exchange matching algorithm, so as to obtain the optimized operation method.

[0065] According to some preferred embodiments of the present invention, the model of the electric carbon calculation system is constructed as follows:

[0066] Intelligent converged terminals and sensing devices form an integration Where u0 represents the intelligent converged terminal, u n n = 1, ..., N represents sensing devices; edge servers interconnected with intelligent converged terminals form a set. The system undergoes T optimization iterations, i.e., the iteration set T = {1,...,t,...,T}. In each iteration, the intelligent fusion terminal senses the topology information of the current iteration based on the electrical data of the sensing devices. Based on the sensed topology information, it formulates a timing data packet delivery strategy. At the same time, it uses the topology information to collect the electrical carbon information obtained by the sensing devices and formulates a strategy for offloading tasks from the sensing devices and allocating computing resources to the edge servers. In this process, the topology state between the intelligent fusion terminal, the sensing devices, and the edge servers remains unchanged in each iteration, but changes in the topology state in different iterations.

[0067] According to some preferred embodiments of the present invention, the time synchronization data packet transmission model is constructed as follows:

[0068] In the t-th iteration, the intelligent fusion terminal or the time-synchronized device u n n = 0, 1, ..., N pairs of unsynchronized devices u n' If n' = 1, 2, ..., N completes the time synchronization, then the time synchronization function x n→n',t =1, otherwise x n→n',t =0;

[0069] Furthermore, in the t-th iteration, the intelligent fusion terminal or the time-synchronized device u n to unsynchronized devices u n′ The transmission rate R of the time synchronization data packets n→n′，t for:

[0070] R n→n',t =N s log2(1+SINR n→n',t (1)

[0071] in,

[0072] Where, N sSINR is the OFDM symbol transmission rate. n→n',t For device u n via path Send data packets to device u n' Signal-to-interference-plus-noise ratio (SIR) at time; p n For u n Transmission power, H represents path gain. n→n',t ζ t Γ and N0 represent the path frequency response, electromagnetic interference, and noise power, respectively. n→n',t This indicates the signal-to-noise ratio gap.

[0073] According to some preferred embodiments of the present invention, the time synchronization error model is constructed as follows:

[0074] The time synchronization delay of the sensing device is:

[0075]

[0076]

[0077] in, τ0 represents the time synchronization delay of the sensing device, τ0 represents the time slot length for each iteration, and num represents the time slot. n For device u n Number of time slots that have been granted time, For u n with u n' Transmission delay; A t Timing The size of the time synchronization data packet;

[0078] And the cumulative time synchronization error is:

[0079]

[0080]

[0081] Where, ω n′，t For u in the t-th iteration n' The cumulative timing error, For u in the t-th iteration n' From device u n The cumulative synchronization error of received timing data packets, ω n，t For u n The cumulative synchronization error, For u n Transmit time synchronization data packets to u n The resulting single-slot error.

[0082] According to some preferred embodiments of the present invention, the electrocarbon information transmission model is constructed as follows:

[0083]

[0084] in, In the t-th iteration, the sensing device u n Transmit carbon information to the edge server connected to the smart fusion terminal m The rate of information transmission of carbon during the time; η n→m,t n = 1, 2, ..., N represents the device u n Task unloading decision variable, when u in the t-th iteration n Choose to offload the carbon information to the server connected to the smart fusion terminal. m To calculate the carbon content, η n→m,t =1, otherwise η n→m,t =0.

[0085] According to some preferred embodiments of the present invention, the total time delay model for calculating the electrocarbon is constructed as follows:

[0086]

[0087]

[0088]

[0089]

[0090] This indicates the total calculation delay for the carbon dioxide generator; This indicates the delay in unloading the carbon information. This indicates the calculation delay of the carbon information; For device u n The amount of collected carbon dioxide information data; For edge servers m The number of computation cycles required to process each bit of carbon information; α m→n For edge servers m For device u n Allocated computing resources; ψ m Edge servers for connecting smart converged terminals m Total computing resources on board.

[0091] According to some preferred embodiments of the present invention, the optimization objectives P1 and P2 of the optimization problem are established as follows:

[0092] P1:

[0093] C1:

[0094] C2:

[0095] C3:

[0096] C4:

[0097] In P1, C1 represents the range of values ​​for the timing decision indicator variable; C2 indicates that only one u exists in each iteration. n For device u n' Its timing, C3 indicates when u n with u n' Time synchronization is not possible without topology connectivity; C4 represents the time delay constraint for time synchronization data packet transmission.

[0098] P2:

[0099] C1:

[0100] C2:

[0101] C3:

[0102] C4:

[0103] In P2, C1 represents the range of values ​​for the indicator variable for the task unloading decision of the carbon information system; C2 represents the value range of device u in each iteration. n The carbon information is offloaded to the edge server connected to the smart fusion terminal. m To complete the carbon dioxide calculation; C3 represents s m with u n Without a topology connection, task unloading is not possible; C4 indicates the server s connected to the smart terminal. m The maximum number of simultaneous values ​​can be q. m Each device performs carbon dioxide calculations.

[0104] According to some preferred embodiments of the present invention, the calculation of the carbon information task offloading and computing resource allocation includes:

[0105] Step 1: Obtain the sensor device u n For edge servers m Preference value β n→m,t Based on the preference value β by the intelligent fusion terminal n→m,t Sort all sensors in descending order to obtain a preference list for sensor devices. Wherein, the preference value β n→m,t The reciprocal of the total delay for the electrocarbon is calculated as follows:

[0106]

[0107] Step 2: Obtain a decision through exchange matching calculation, including: initializing device task unloading decision traversal η. n→m,t Preference list Matching relation Λ, utility function Then, the preference value is calculated according to equation (23), and the preference list is updated. Secondly, a temporary matching relationship is established, with the intelligent fusion terminal serving as the device u n A request is sent to the server, when the edge server s m The number of received requests is less than q m At that time, the edge server s m Accept device requests, if edge server s m The number of requests received is greater than q m Then accept the first q. m One request is made, and requests from other devices are rejected, while the matching relationship Λ(s) is updated. m )=u n Further, construct exchange matching relationships based on the established temporary matching relationships. For equipment u n and Reallocate server computing resources and update utility functions. If, after swapping matching relationships, the utility functions of all devices are... If the value increases, the device update matching relationship will be as follows: and Simultaneously update the task uninstallation decision variables and clear the swap matching relationships. Finally, when no exchange matching relationship exists, the matching terminates, completing the intelligent fusion terminal's decision-making for the unloading of carbon information tasks and resource allocation.

[0108] The present invention further provides an intelligent fusion terminal for implementing the above-mentioned carbon dioxide calculation method, comprising an application layer, a physical layer, and a protocol interface layer. The application layer includes a topology-aware container for obtaining the topology information, a time synchronization container for completing the synchronization, and a carbon dioxide calculation container for obtaining the carbon dioxide information task offloading strategy and the carbon dioxide calculation resource allocation scheme. The physical layer supports the operation of the intelligent fusion terminal and the implementation of the functions of the application layer, and connects to multiple edge servers. The protocol interface layer enables information interaction and interactive information processing between the application layer and the physical layer.

[0109] The beneficial effects of the technical solution of this invention are as follows:

[0110] 1. This invention can achieve redundancy-free and accurate BP neural network topology prediction through the topology sensing container of an intelligent fusion terminal. It uses the first layer neural network to identify normal and abnormal data in the electrical data sample group collected by the sensing device, classifies the data samples, further removes redundant and abnormal data and inputs it into the BP neural network, avoids the impact of abnormal data on convergence performance, realizes accurate and low-cost precise topology sensing of smart parks, and solves the problem that the inaccurate topology sensing of existing topology identification methods leads to the lack of accurate topology information in the timing data packet transmission and the electric carbon information task offloading strategy.

[0111] 2. After obtaining the topology connection information, this invention optimizes the timing data packet transmission decision by using a time synchronization container to carry an efficient DQN based on Q-difference. In some preferred embodiments, the decision fully considers the impact of campus topology information, device time synchronization errors, and electromagnetic interference on the timing data packet transmission delay. With the goal of minimizing the average error of the time-synchronized device under transmission delay constraints, the efficient DQN learning capability based on Q-difference is utilized to enable the intelligent fusion terminal to formulate a timing data packet transmission strategy for the device, thereby improving the time synchronization accuracy of the smart campus.

[0112] 3. In some specific embodiments, the present invention further establishes a model for the transmission latency and computation latency of carbon computing. By using a carbon computing container to carry an exchange matching algorithm, and by considering the impact of different task offloading strategies on the allocation of computing resources, an exchange matching relationship is set. The matching relationship is adjusted by adjusting the total carbon computing latency, and a low-latency task offloading strategy and computing resource allocation strategy are formulated to reduce the total carbon computing latency. This enables efficient transmission and computation of carbon information of smart park equipment, supporting the efficient and stable operation of low-carbon services.

[0113] 4. This invention integrates topology sensing, time synchronization, offloading strategy optimization, and electric carbon calculation functions into a single intelligent fusion terminal by deploying a topology sensing container, a time synchronization container, and an electric carbon calculation container in the intelligent fusion terminal. This strengthens the connection between the functions, reduces communication overhead between terminals, and enables low-cost and accurate topology sensing, high-precision and fast time synchronization, and low-latency electric carbon calculation in smart parks. Attached Figure Description

[0114] Figure 1 This is a structural diagram of the intelligent fusion terminal for electrocarbon computing based on edge containers according to the present invention;

[0115] Figure 2 This is a schematic diagram of the overall principle of the optimization algorithm for topology sensing, time synchronization, and carbon dioxide calculation in the intelligent fusion terminal of the smart park in this invention.

[0116] Figure 3 This is an application scenario diagram of the intelligent fusion terminal for smart park carbon calculation in this invention;

[0117] Figure 4 This is a schematic diagram of the redundancy-removing, precise BP neural network topology sensing principle of the present invention;

[0118] Figure 5 This is a schematic diagram of the efficient DQN time synchronization and timing data packet transmission routing optimization method based on Q difference of the present invention.

[0119] Figure 6 This is a flowchart of the method for optimizing the unloading and resource allocation of electric carbon computing tasks based on exchange matching according to the present invention. Detailed Implementation

[0120] The present invention will now be described in detail with reference to embodiments and accompanying drawings. However, it should be understood that the embodiments and drawings are for illustrative purposes only and do not constitute any limitation on the scope of protection of the present invention. All reasonable modifications and combinations included within the inventive spirit of the present invention fall within the scope of protection of the present invention.

[0121] See attached document Figure 1 According to the technical solution of the present invention, a specific implementation structure of the edge container-based intelligent fusion terminal for carbon computing includes:

[0122] The application layer contains a topology-aware container, a time synchronization container, and an electric carbon calculation container. The topology-aware container is used to achieve accurate topology perception, the time synchronization container is used to perform time synchronization processing based on the topology information obtained from the topology-aware container, and the electric carbon calculation container is used to perform electric carbon calculation based on the time synchronization information obtained from the time synchronization container.

[0123] A physical layer containing multiple hardware modules that supports the operation of intelligent converged terminals and enables the application layer to implement its functions;

[0124] This is a protocol interface layer that enables information interaction and processing between the application layer and the physical layer, such as obtaining sensor data or issuing sensor transmission task offloading strategies through the physical layer, preprocessing the obtained sensor data, providing data support for the operation of each container in the application layer, and realizing communication and coordination between each container.

[0125] Preferably, each container in the application layer implements its corresponding function by carrying out an optimized operation method.

[0126] Preferably, the physical layer includes the following hardware modules: a power module, a CPU module, a communication module, a storage module, a computing module, and a control module. The power module supplies power to the intelligent converged terminal; the CPU module processes data and control commands; the communication module exchanges information with other devices via power line carrier; the storage module stores data acquired or generated by the converged terminal; the computing module supports data processing and computation for each container in the application layer; and the control module coordinates the functions between the various containers.

[0127] Preferably, the protocol interface layer includes a protocol unit that provides data support and a communication unit that provides internal communication. The protocol unit includes a power line carrier communication interface that is equipped with a power-specific protocol and a communication protocol, and the communication unit includes an internal bus interface containing a data bus and a control bus.

[0128] Preferably, providing data support includes: receiving and preprocessing information collected by the sensing device, and transmitting task offloading decisions to the sensing device.

[0129] Preferably, providing internal communication includes: enabling communication and coordination between containers in the application layer. In this preferred embodiment, the internal bus interface connects the containers in the application layer through a control bus and a data bus.

[0130] Preferably, the dedicated power protocol is selected from dedicated power protocols such as IEC104 or IEC61850.

[0131] Preferably, the communication protocol is selected from communication protocols such as MQTT or CoAP.

[0132] Further, refer to the appendix. Figure 2 Regarding the aforementioned fusion terminal or its preferred embodiments, this invention provides a method for achieving low-cost, accurate topology sensing, high-precision, fast time synchronization, and low-latency carbon dioxide calculation. It also provides specific implementation procedures for this method in application scenarios of carbon dioxide calculation in smart parks, including:

[0133] S1 obtains the topology information of the smart park through the topology sensing model via the topology sensing container based on the device information and electrical data of each sensing device in the smart park obtained from the protocol interface layer. The topology information includes channel state information and a topology connection matrix containing the connection relationship between the fusion terminal, sensing devices and edge servers.

[0134] S2, based on the obtained topology information, uses the time synchronization container to perform time synchronization processing on each device in the smart park through an optimized operation method, so that each device has time consistency.

[0135] After achieving time consistency, the S3 electric carbon computing container obtains an electric carbon information task unloading strategy and an electric carbon computing resource allocation scheme through an optimized operation method. The sensing device unloads the computing information according to the electric carbon information task unloading strategy, and the edge server completes the allocated computing resources according to the electric carbon computing resource allocation scheme to obtain low-latency electric carbon computing results.

[0136] In S1, preferably, refer to the attached document. Figure 4 The topology-aware model is constructed based on a two-layer neural network consisting of a SOM network and a BP network, and specifically includes: a first-layer neural network formed by the SOM network to achieve effective data classification and filtering, and a second-layer neural network formed by the BP network to perform topology awareness based on the effective data obtained from the first-layer neural network; wherein, the SOM network includes an input layer and a competition layer, i.e., its output layer, and the BP network includes a hidden layer and an output layer.

[0137] Under this topology-aware model, the topology-aware container can effectively remove redundant information and obtain accurate topology information.

[0138] Furthermore, under this topology-aware model, referring to the appendix... Figure 3 Its topology sensing process includes:

[0139] The first layer of the neural network is trained based on a historical electrical database data sample group containing labels for normal data sample groups and abnormal data sample groups.

[0140] All electrical data samples received from the device in the current time slot are input into the first layer of the trained neural network for classification, and data classified as abnormal sample groups are removed.

[0141] The data classified as normal data sample group is input into the second layer neural network to perceive the topological connection relationship under the current time slot, and the accuracy of the topological perception of the second layer neural network is continuously adjusted and improved until a high-accuracy second layer neural network is obtained.

[0142] The obtained high-precision second-layer neural network outputs topological information based on the data from the normal data sample group.

[0143] More preferably, the classification is performed by the SOM neural network by selecting the winning neuron in the competitive layer and continuously updating the weight parameters and data radius, specifically including:

[0144] (1) In the current iteration, the Euclidean distance d with the input layer neuron is determined. i,j The winning neuron in the competitive layer is determined by minimizing the Euclidean distance, which is calculated using the following formula:

[0145]

[0146] Among them, w i,j (e) represents the connection weight between input layer neuron i and competing layer neuron j in the current e-th iteration; y1(e) represents the electrical data sample group y n The input vector Y = (y1(e),...,y2(e),...,y) = <ν,θ> is (t) = <ν,θ>. n (e)) T In the i-th element, ν represents the sample group of equipment operating voltage phase angle data, θ represents the sample group of equipment operating injected power data, and e represents the set of iterations when the SOM neural network performs classification: E = (1,...,e,...,E). T In the context of the e-th iteration, E represents the total number of iterations, and T represents the transpose.

[0147] (2) After obtaining the winning neuron, the connection weights between the input layer neuron i and the competing layer neuron j are continuously adjusted during the iteration, as follows:

[0148] w i,j (e+1)=w i,j (e)+χ(e)h(e)(y i,t -w i,j (e)) (14)

[0149] Where 0 < χ(e) < 1 is the learning rate function, and h(e) is the neighborhood function, the value of which decays over time; preferably, the learning rule is set as follows:

[0150]

[0151]

[0152]

[0153] Where, d j,j' Let be the distance between competing layer neuron j and competing layer neuron j', χ(0) be the initial learning rate, and r(e+1) be the data radius. This is the floor function.

[0154] (3) In subsequent iterations, let e = e + 1, and repeat the above iteration process until the neural network learning rate drops to 0, obtain the data radius of all equipment electrical data sample groups, and set the radius constraint r. min Because the abnormal electrical data samples fluctuate significantly, their classified data radius is larger than that of the normal electrical data samples. Therefore, they will exceed the radius constraint r. min The data sample groups were removed to obtain the normal data sample groups.

[0155] More preferably, obtaining the output topology information includes:

[0156] (1) Input the normal electrical data sample group after the SOM neural network classification and elimination into the hidden layer of the BP neural network;

[0157] (2) The network relationship is continuously updated using the actual input and expected output of the BP neural network to obtain the perceived channel state information. and topology connection matrix ,in, This includes the connection relationships between intelligent converged terminals and devices, as well as the connection relationships between edge servers and devices, as follows:

[0158]

[0159] Where N0 represents channel noise power information, ζ represents channel electromagnetic interference information, and Z... n→n',t Z is the indicator variable for the topology connection between devices. n→m,t For devices and edge servers m Inter-topology connection indicator variable.

[0160] Furthermore, in S2, refer to the appendix. Figure 5 The time synchronization container is implemented based on a DQN (deep Q-network) network, and the time synchronization process is completed through an efficient DQN algorithm based on Q-difference, specifically including:

[0161] Based on the empirical data derived from the Q-difference, empirical data is preferentially extracted from the empirical pool according to the probability.

[0162] Based on the extracted empirical data, deep reinforcement learning of neural networks based on the DQN architecture is carried out through the DQN iterative algorithm.

[0163] The timing data packet transmission strategy is output by the neural network based on the DQN architecture. The timing data packet transmission strategy selects a timed device with a topological link and a small cumulative timing error to send timing data packets to a timed device that has not been timed.

[0164] The devices within the smart park synchronize their time by following the time synchronization data packet transmission strategy.

[0165] The state space, action space, and reward function in the DQN iterative algorithm are defined as follows:

[0166] State space: Defined by the size of the timing data packets obtained in each iteration. and the topology connection matrix The composition is as follows:

[0167]

[0168] Action Space: Defined as the set of timing data packet transmission strategies formulated by the intelligent fusion terminal for the device, as follows:

[0169]

[0170] Among them, u n Indicates that the time synchronization device, u n' Indicates a device that has not received a time synchronization signal; Z represents a collection of intelligent converged terminals and devices; n→n',t Indicates the topology connection indicator variable between devices; x n→n',t For the t-th iteration, the intelligent fusion terminal or the time-synchronized device u n n = 0, 1, ..., N pairs of unsynchronized devices u n' The timing completion value for n' = 1, 2, ..., N is specifically: when the t-th iteration occurs, the non-timing device u n' Obtain intelligent converged terminal or time-synchronized device u n If it is granted, then x n→n',t =1, otherwise x n→n',t =0.

[0171] Reward function: The reward function is defined as the negative of the cumulative error in device time synchronization, as follows:

[0172]

[0173] in, For the untimed device u in the t-th iteration n' From the time-synchronized device u n The cumulative synchronization error of the received time synchronization data packets satisfies:

[0174]

[0175] The first term is u n The cumulative synchronization error, the second term is u n Transmit time synchronization data packets to u n' The resulting single-slot error.

[0176] The deep reinforcement learning includes: obtaining the state in each iteration through a neural network based on the DQN architecture. Next, take action n',t The Q value, i.e., the estimated value of Q. Subsequently, by continuously updating the neural network parameters, the estimated value of the obtained Q-value is improved. By approximating the actual value, a neural network is finally obtained that can accurately output actions under state input by fitting the "state-action" relationship.

[0177] Unlike traditional DQN algorithms that use random sampling to extract data from the experience pool, resulting in poorly performing data being reused for network updates, leading to decreased neural network fitting accuracy and slow convergence, the efficient DQN algorithm based on Q-difference provided in this invention sets up an experience pool based on Q-difference. Based on this experience pool, the following priority extraction probabilities for empirical data based on Q-difference were set:

[0178] Record the estimated Q value of each set of experience data in the experience pool, and compare the difference between the estimated Q value of the experience data and the Q value of the target network. The smaller the difference, the higher the extraction priority and the greater the extraction probability.

[0179] This method ensures that data beneficial to network update speed is extracted first, improving the convergence speed of the DQN network and enabling it to converge to the optimal strategy more quickly.

[0180] Furthermore, the deep reinforcement learning process includes:

[0181] Step 1: Network initialization, including: initializing the evaluation network parameters, target network parameters, and experience pool;

[0182] Step 2: At the beginning of each iteration, use the ε-greedy algorithm to assign a time slot number to device u in the current time slot. n' The timing data packet transmission strategy is formulated, which specifically includes: the intelligent fusion terminal randomly explores with probability ε and randomly selects an action, or uses the action with the largest estimated Q value with probability 1-ε. n',t ;

[0183] Step 3: The device completes device timing according to the timing data packet transmission strategy formulated by the intelligent fusion terminal, and calculates the reward value Reward according to formula (5) and formula (21). n',t Then transition to the next state S n',t+1 At the same time, experience data is generated. Store in the experience pool;

[0184] Step 4: Extract experience data from the experience pool according to the priority extraction probability of the experience data. The priority extraction probability of the experience data is set as follows:

[0185]

[0186] in, For the estimated Q value of the k-th set of empirical data, Let Q be the target network's Q value. Since empirical data with a smaller difference from the target network's Q value perform better for network training, this formula implies that the smaller the difference from the target network's Q value, the higher the extraction probability.

[0187] Step 5: Based on the extracted empirical data, calculate the loss function of the neural network, update the parameters of the valley network and the target network, and repeat steps 2 to 5 until the optimization is completed after T iterations.

[0188] Furthermore, in some specific embodiments, obtaining the optimized operation method includes:

[0189] (1) Constructing an electric carbon calculation system model

[0190] See attached document Figure 3 The illustration depicts an application scenario of a smart park electricity carbon calculation intelligent fusion terminal. It comprises one intelligent fusion terminal connected to M edge servers and N sensing devices. The intelligent fusion terminal obtains clock information via 5G time synchronization and BeiDou satellite connectivity, serving as a clock source for timing the sensing devices. The edge servers, equipped with computing resources, provide computing services to the intelligent fusion terminal. The sensing devices collect electricity carbon information and, in response to the intelligent fusion terminal's decisions, offload this information to the edge servers to complete the park's electricity carbon calculation.

[0191] In the above application scenarios, the following electric carbon calculation system model can be constructed:

[0192] Intelligent converged terminals and sensing devices form an integration Where u0 represents the intelligent converged terminal, u n n = 1, ..., N represents the sensing devices;

[0193] Define and connect edge servers that are interconnected with intelligent converged terminals to form a set.

[0194] The system undergoes T optimization iterations, i.e., the iteration set T = {1,...,t,...,T}. In each iteration, the intelligent fusion terminal senses the topology information of the current iteration based on the electrical data of the sensing devices. Based on the sensed topology information, it formulates a timing data packet delivery strategy. At the same time, it uses the topology information to collect the carbon information obtained by the sensing devices and formulates a strategy for offloading tasks from the sensing devices and allocating computing resources to the edge servers, thereby achieving low-latency carbon calculation. In this process, it is set that the topology state between the intelligent fusion terminal, the sensing devices, and the edge servers remains unchanged in each iteration, but the topology state changes in different iterations.

[0195] (2) Constructing a time-synchronized data packet transmission model

[0196] In power line carrier networks, sensing devices serve both as timing target devices and, after completing their own timing, as new clock sources to provide timing for other devices. Therefore, intelligent converged terminals formulate reasonable timing data packet transmission strategies for devices, directly providing timing for connected sensing devices, and also providing timing for unconnected sensing devices through the transmission of timing data packets between sensing devices.

[0197] Define x n→n',t For the t-th iteration, the intelligent fusion terminal or the time-synchronized device u n n = 0, 1, ..., N pairs of unsynchronized devices u n' If, in the t-th iteration, the untimed device u is not timed, and n' = 1, 2, ..., N, time synchronization is completed. n' Through intelligent converged terminals or time-synchronized devices n If the time is given, then x n→n',t =1, otherwise x n→n',t =0.

[0198] Preferably, considering the complex electromagnetic environment and noise interference in the smart park, this invention uses orthogonal frequency division multiplexing (OFDM) technology to achieve time synchronization data packet transmission. Assuming that each iteration involves the intelligent fusion terminal sending data packets to all sensing devices, the transmission rate of the time synchronization data packets is...

[0199] R n→n',t =N s log2(1+SINR n→n',t (1)

[0200] Where, N s SINR is the OFDM symbol transmission rate. n→n',t For device u n via path f un→un' Send data packets to device u n' Signal-to-interference-to-noise ratio at that time.

[0201] SINR n→n',t Represented as:

[0202]

[0203] Where, p n For u n Transmission power, H represents path gain. n→n',t ζ t Γ and N0 represent the path frequency response, electromagnetic interference, and noise power, respectively. n→n',t This indicates the signal-to-noise ratio gap.

[0204] (3) Constructing a time synchronization error model

[0205] Each iteration is divided into time slots of length τ0. At the beginning of each time slot, the intelligent fusion terminal or the sensing device that has completed time synchronization transmits time synchronization data packets to the sensing devices that have not yet completed time synchronization. During the transmission of data packets across multiple time slots, the time synchronization delay of the sensing devices is determined by the device u. n Time synchronization delay and device u completed n With device u n' The transmission delay consists of two parts, represented as:

[0206]

[0207] Where, num n For device u n Number of time slots that have been granted time, For u n with u n' The transmission delay can be calculated as follows:

[0208]

[0209] A t Timing This refers to the size of the time synchronization data packet. To ensure the timeliness of the time synchronization data packet transmission, preferably, the present invention imposes constraints on the transmission delay, namely...

[0210] Define u in the t-th iteration n' The cumulative time synchronization error is in, For u in the t-th iteration n' From device u n The cumulative synchronization error of the received time synchronization data packets can be expressed as:

[0211]

[0212] The first term is u n The cumulative synchronization error, the second term is u n Transmit time synchronization data packets to u n' The resulting single-slot error.

[0213] (4) Constructing a carbon information transmission model for equipment

[0214] Because the uploading of carbon information by sensing devices is affected by both the transmission topology and the computing resources of the edge servers providing computing services to intelligent fusion terminals, different offloading schemes for the carbon information from the devices result in differences between the computation latency of carbon calculation and the information transmission latency. Define η. n→m,t n = 1, 2, ..., N represents the device u n Task unloading decision variable, when u in the t-th iteration nChoose to offload the carbon information to the server connected to the smart fusion terminal. m To calculate the carbon content, η n→m,t =1, otherwise η n→m,t =0.

[0215] Therefore, in the t-th iteration, the sensing device u n Transmit carbon information to the server connected to the intelligent fusion terminal m The data transmission rate of carbon information at that time is

[0216]

[0217] The meanings of each symbol are the same as those in the time synchronization data packet.

[0218] (5) Constructing a total time delay model for calculating carbon dioxide

[0219] The total latency for calculating carbon emissions at the intelligent fusion terminal includes two parts: the latency for unloading carbon emissions information from the device and the latency for calculating carbon emissions information. (Carbon emissions information unloading latency) It can be represented as:

[0220]

[0221] in, For device u n The size of the collected carbon information data.

[0222] The server u connected to the intelligent converged terminal n Calculation delay of carbon dioxide information It can be represented as:

[0223]

[0224] in, For server s m The number of computation cycles required to process each bit of carbon information, α m→n For server s m For device u n The allocated computing resources. Due to the varying amounts of carbon information data across different devices, the server... m The allocation of computing resources to each device differs, where the resource allocation for device u is... n The allocated computing resources can be represented as:

[0225]

[0226] Where, ψ m The server connected to the intelligent converged terminal m The total computing resources used. Therefore, the total latency of carbon-based computing can be expressed as:

[0227]

[0228] (6) Model the optimization problem

[0229] The realization of high-precision and fast time synchronization and low-latency carbon calculation in smart parks requires the support of park topology information. However, smart parks are affected by complex electrical environments. The random access and switching of power loads in the network, the resulting changing impedance and unpredictable strong noise interference, can easily lead to poor power line carrier communication and cause changes in topology information.

[0230] Therefore, this invention first uses a smart fusion terminal topology sensing container to sense the topology information of the park through a low-cost and accurate topology sensing method, and then uses the sensed topology information to solve the problems of time synchronization and timing data packet transmission optimization and electric carbon computing task unloading and resource allocation optimization.

[0231] The optimization problem for time synchronization and timing data packet transmission is constructed as follows:

[0232] This invention addresses the time synchronization data packet transmission optimization problem by constructing a time synchronization container that minimizes the average cumulative time synchronization error in each iteration, P1. This optimization objective is expressed as:

[0233] P1:

[0234] C1:

[0235] C2:

[0236] C3:

[0237] C4:

[0238] In P1, C1 represents the range of values ​​for the timing decision indicator variable; C2 indicates that only one u exists in each iteration. n For device u n' Its timing, C3 indicates when u n with u n' Time synchronization is not possible without topology connectivity; C4 is a time delay constraint for time synchronization data packet transmission to ensure the reliability of time synchronization data packets.

[0239] The optimization problem of offloading and allocating resources for carbon computing tasks is constructed as follows:

[0240] This invention addresses the optimization problem of task unloading and resource allocation in carbon computing by minimizing the total latency P2 of carbon computing in the carbon computing container. This is expressed as:

[0241] P2:

[0242] C1:

[0243] C2:

[0244] C3:

[0245] C4:

[0246] In P2, C1 represents the range of values ​​for the indicator variable for the task unloading decision of the carbon information system; C2 represents the value range of device u in each iteration. n The carbon information is offloaded to the server connected to the smart fusion terminal. m To complete the carbon dioxide calculation; C3 represents s m with u n Without a topology connection, task unloading is not possible; C4 indicates the server s connected to the smart terminal. m The maximum number of simultaneous values ​​can be q. m Each device performs carbon dioxide calculations.

[0247] (7) The above optimization problem is solved by using an exchange-match-based optimization method for unloading and allocating electric carbon computing tasks in the electric carbon computing container.

[0248] The specific steps are as follows:

[0249] Define exchange matching as Where S and The two parties involved in the matching process are the set of servers connected to the intelligent fusion terminal and the set of sensing devices. For the sensor device preference list;

[0250] Define based on preference list The matching relationship is Λ, representing the set. With sets The mapping, Λ(s) m )=u n Indicates server s m With device u n A matching relationship has been established, and Λ(u) n ) = s m ;

[0251] Define the current time and server u n The set of devices for establishing matching relationships is Λ U ;

[0252] Since the device can choose different servers to offload tasks, and each server can provide carbon dioxide calculation services to multiple devices simultaneously, when different sensing devices offload carbon dioxide information to the server, the difference in the amount of carbon dioxide information data will lead to differences in the allocation of computing resources, further affecting the total latency of carbon dioxide calculation. In other words, server resources are affected by the devices offloading carbon dioxide information. Therefore, a utility function is defined. To describe the utility value of the currently formed matching relationship;

[0253] definition The exchange matching relationship in the exchange matching refers to the currently existing matching relationship. The matching objects can be swapped, that is, the matching relationship can be changed. Meanwhile, the matching relationships between other devices and the server remain unchanged. Furthermore, if the utility function increases after the matching relationship changes, then the exchange relationship holds.

[0254] Based on the above definitions, a preference list is constructed and exchange matching calculations are performed to complete the intelligent fusion terminal's decision-making for the unloading of carbon information tasks and resource allocation.

[0255] Furthermore, the process for constructing and exchanging the preference list for matching calculations is as follows: Figure 6 As shown, it specifically includes:

[0256] Step 1: Constructing the preference list. This invention defines that at the t-th iteration, device u... n For server s m Preference value β n→m,t The reciprocal of the total delay for calculating the carbon dioxide is expressed as:

[0257]

[0258] Intelligent converged terminals based on preference value β n→m,t Sort all sensors in descending order to obtain a preferred list of devices. Complete the construction of the preference list.

[0259] Step 2: Exchange and Matching. First, initialization is performed, initializing the device task offloading decision traversal η. n→m,t Preference list Matching relation Λ, utility function Then, calculate the preference value according to formula (23) and update the preference list. Secondly, a temporary matching relationship is established, with the intelligent fusion terminal serving as the device u n Send a request to the server, when the server s m The number of received requests is less than q m At that time, the server s m Accept device requests, if server sm The number of requests received is greater than q m Then accept the first q. m One request is made, and requests from other devices are rejected, while the matching relationship Λ(s) is updated. m )=u n Further, construct exchange matching relationships based on the established temporary matching relationships. For equipment u n and Reallocate server computing resources and update utility functions. If, after swapping matching relationships, the utility functions of all devices are... If the value increases, the device update matching relationship will be as follows: and Simultaneously update the task uninstallation decision variables and clear the swap matching relationships. Finally, when no exchange matching relationship exists, the matching terminates, completing the intelligent fusion terminal's decision-making for the unloading of carbon information tasks and resource allocation.

[0260] In the above optimized operation method, the sensing device can forward the time synchronization data packet through the power line carrier. When the intelligent fusion terminal broadcasts the time synchronization data packet, it uses the network topology information provided by the topology sensing container through the time synchronization container to minimize the cumulative time synchronization error and select an appropriate timing strategy to complete the time synchronization of the intelligent fusion terminal of the smart park with the sensing device.

[0261] In summary, the above-mentioned carbon dioxide calculation method of this invention fully utilizes the advantages of edge container technology, such as independence from the operating environment, ease of deployment on terminals, and low application resource overhead, making it suitable for deployment in smart park smart fusion terminal applications where storage and computing resources are limited. It designs an edge computing-based smart fusion terminal for carbon dioxide calculation and a time synchronization method. By deploying topology sensing containers, time synchronization containers, and carbon dioxide calculation containers on the fusion terminal using edge container technology, and by deploying optimized algorithms within these containers, it achieves low-cost and accurate topology sensing, high-precision and fast time synchronization, and low-latency carbon dioxide calculation in smart parks, supporting the stable and efficient operation of time synchronization and carbon dioxide calculation in smart parks.

[0262] Furthermore, according to embodiments of this disclosure, the methods described above can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program tangibly embodied on a readable medium thereof, the computer program containing program code for performing the methods described above.

[0263] According to embodiments of this disclosure, a single computer architecture as described above can be used to implement the method according to embodiments of this disclosure, or multiple computer architectures as described above can be used in cooperation to implement the method according to embodiments of this disclosure.

[0264] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0265] The units or modules described in the embodiments of this disclosure can be implemented in software or programmable hardware. The described units or modules can also be located in a processor, and the names of these units or modules do not necessarily constitute a limitation on the unit or module itself.

[0266] In another aspect, this disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the apparatus described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores one or more programs that are used by one or more processors to perform the methods described in this disclosure.

[0267] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A method for synchronizing the time of electrocarbon computing based on an intelligent fusion terminal with edge containers, characterized in that, It includes: S1 Obtain the system topology information, wherein the topology information includes channel state information and a topology connection matrix containing the connection relationships between the intelligent fusion terminal, sensing devices and edge servers; S2 Based on the topology information, the system synchronizes the timing of each sensing device by optimizing the operation method, so that each sensing device has time consistency. After achieving time consistency, S3 obtains the carbon information task offloading strategy and the carbon computing resource allocation scheme. The sensing device offloads the computing information according to the carbon information task offloading strategy, and the edge server completes the allocated computing resources according to the carbon computing resource allocation scheme to obtain low-latency carbon computing results. The optimized operation method is obtained by solving the following optimization problems: (1) minimizing the average time synchronization cumulative error of each iteration; (2) minimizing the total time delay of the electric carbon calculation; the topology information is obtained by a two-layer neural network composed of a SOM neural network and a BP neural network, the two-layer neural network including: a first-layer neural network formed by the SOM neural network to realize effective data classification and screening and a second-layer neural network formed by the BP neural network to perform topology perception based on the effective data obtained by the first-layer neural network; wherein, the SOM neural network includes an input layer and a competition layer, wherein the output layer of the SOM neural network is used as the competition layer; the BP neural network includes a hidden layer and an output layer; the process of obtaining the topology information through the two-layer neural network includes: The first layer of the neural network is trained based on a historical electrical database data sample group containing labels for normal data sample groups and abnormal data sample groups. All electrical data samples from the sensing devices received in the current time slot are input into the first layer of the trained neural network for classification, and data classified as abnormal sample groups are removed. The data classified as normal data sample group is input into the second layer neural network to perceive the topological connection relationship under the current time slot, and the accuracy of the topological perception of the second layer neural network is continuously adjusted and improved until a high-accuracy second layer neural network is obtained. The obtained high-precision second-layer neural network outputs topological information based on the data from the normal data sample group; The classification includes: (1) In the current iteration of the SOM neural network, the winning neuron in its competitive layer is determined by minimizing the Euclidean distance between itself and the neurons in its input layer, wherein the Euclidean distance is calculated using the following formula: ， in, This indicates the number of input layer neurons in the current e-th iteration. Competing layer neurons Connection weights between them; Indicates electrical data sample group input vector The i-th element in This represents a sample group of phase angle data for the operating voltage of the equipment. This represents a sample group of injected power data during device operation, and e represents the set of iterations performed by the SOM neural network for classification. In the e-th iteration, The total number of iterations is represented by T, and the transpose is represented by T. (2) After obtaining the winning neuron, the input layer neurons are continuously adjusted during the iteration. Competing layer neurons The connection weights between them are as follows: ， in, The learning rate function; It is a neighborhood function related to the data radius, and its value decays over time; Repeat the iterative process until the neural network learning rate drops to 0, obtaining the data radius of the electrical data sample group of all sensing devices, and then removing those exceeding the data radius constraint. Data sample groups were marked as outliers and removed.

2. The method for synchronizing the calculation time of carbon dioxide according to claim 1, characterized in that, in, The acquisition of the output topology information includes: The normal electrical data sample group after being classified and eliminated by the SOM neural network is input into the hidden layer of the BP neural network; By continuously updating the network relationship using the actual input and desired output of the BP neural network, the perceived channel state information is obtained. and topology connection matrix ,in, This includes the connection relationships between intelligent converged terminals and devices, as well as the connection relationships between edge servers and devices, as follows: ， in, For channel noise power information, For channel electromagnetic interference information, This serves as an indicator variable for the topology connections between sensing devices. For sensing devices and edge servers Inter-topology connection indicator variable.

3. The method for synchronizing the calculation time of carbon dioxide according to claim 1, characterized in that, The time synchronization is achieved through a DQN network using an efficient DQN algorithm based on Q-difference, which specifically includes: Based on the empirical data derived from the Q-difference, empirical data is preferentially extracted from the empirical pool according to the probability. Based on the extracted empirical data, deep reinforcement learning of neural networks based on the DQN architecture is carried out through the DQN iterative algorithm. The timing data packet transmission strategy is output by the neural network based on the DQN architecture. The timing data packet transmission strategy includes selecting timed devices with existing topology links and small cumulative timing errors for untimed devices to send timing data packets to them. Each sensing device completes device time synchronization according to the time synchronization data packet transmission strategy; The deep reinforcement learning includes: obtaining the state in each iteration through a neural network based on the DQN architecture. Take action below The Q value, i.e., the estimated value of the Q value. Subsequently, by continuously updating the neural network parameters, the estimated value of the obtained Q-value is improved. By approximating the actual value, a neural network capable of achieving precise action output under given state input is ultimately obtained; among which, This represents the set of time-synchronized data packet transmission strategies formulated by the intelligent fusion terminal for the sensing device; Among them, the empirical data priority extraction probability based on Q difference The settings are as follows: ， in, For the estimated Q value of the k-th set of empirical data, Let Q be the target network Q-value.

4. The method for synchronizing the calculation time of carbon dioxide according to claim 3, characterized in that, The state space, action space, and reward function in the DQN iterative algorithm are set as follows: State space: ， in, Represents the topology connection matrix. This indicates the size of the timing data packet obtained in each iteration; Action space: ， in, This represents the set of time-synchronized data packet transmission strategies formulated by the intelligent fusion terminal for the sensing device. This indicates that the time synchronization device has been activated. Indicates a device that has not received a time synchronization signal; It represents a collection of intelligent converged terminals and devices; Indicates the topology connection indicator variable between devices; For the first In the next iteration, intelligent converged terminals or time-synchronized devices For devices without time synchronization The timing completion value is specifically: when the first... In the next iteration, the time-unsynchronized device Obtain intelligent converged terminals or time-synchronized devices If it provides the time, then Conversely, ; Reward function: ， in, For the first Devices without timing in the next iteration From the time-synchronized device The cumulative synchronization error of the received time synchronization data packets satisfies: ， The first item is The cumulative synchronization error, the second term is Transmit time synchronization data packets to The resulting single-slot error.

5. The method for synchronizing the calculation time of carbon dioxide according to claim 1, characterized in that, The optimized operation method is obtained by: (1) Construct an electric carbon computing system model containing the intelligent fusion terminal, the edge server, the sensing device and their interconnections; (2) Construct a timing data packet transmission model based on the synchronization timing strategy under the electric carbon computing system model; (3) Based on the time synchronization data packet transmission model, construct a time synchronization error model under the electric carbon computing system model; (4) Based on the time-synchronization data packet transmission model, construct the electric carbon information transmission model under the electric carbon calculation system model; (5) Based on the time synchronization error model and the electric carbon information transmission model, construct the total delay model of electric carbon calculation for the electric carbon calculation system model; (6) Based on the electric carbon computing system model, the time synchronization data packet transmission model, the time synchronization error model, the carbon information transmission model, and the electric carbon computing total delay model, the optimization problem is set, and the optimization problem is solved by the electric carbon information task unloading and computing resource allocation calculation based on the exchange matching algorithm, so as to obtain the optimized operation method.

6. The method for synchronizing the calculation time of carbon dioxide according to claim 5, characterized in that, in, The model for the carbon dioxide calculation system is constructed as follows: Intelligent converged terminals and sensing devices form an integration ,in Representing intelligent converged terminals, Representative sensing devices; edge servers interconnected with intelligent converged terminals form a collection. The system has conducted a total of Sub-optimization iteration, i.e., iteration set In each iteration, the intelligent fusion terminal senses the topology information of the current iteration based on the electrical data of the sensing devices. Based on the sensed topology information, it formulates the timing data packet delivery strategy. At the same time, it uses the topology information to collect the electrical carbon information obtained by the sensing devices and formulates the sensor device task offloading and edge server computing resource allocation strategy. In this process, the topology state between the intelligent fusion terminal, the sensing devices and the edge server remains unchanged in each iteration, but its topology state changes in different iterations. And / or, the timing data packet transmission model is constructed as follows: No. In the next iteration, intelligent converged terminals or time-synchronized devices For devices without time synchronization Once time synchronization is complete, the time synchronization function will be used. Conversely, ; Furthermore, in the t-th iteration, the intelligent fusion terminal or the time-synchronized device... To unsynchronized devices The transmission rate of the time synchronization data packets for: ， in, , in, For OFDM symbol transmission rate, For equipment via path Send data packets to the device Signal-to-interference-to-noise ratio (SIR) at that time; for Transmission power, Indicates path gain. , and These represent path frequency response, electromagnetic interference, and noise power, respectively. Indicates the signal-to-noise ratio gap; And / or, the time synchronization error model is constructed as follows: The time synchronization delay of the sensing device is: ， ， in, This indicates the time synchronization delay of the sensing device. This indicates the length of the time slot to be divided in each iteration. For equipment Number of time slots that have been granted time, for and Transmission delay; The size of the time synchronization data packet; And the cumulative time synchronization error is: ， ， in, For the first In the next iteration The cumulative time synchronization error, For the first In the next iteration From the equipment The cumulative synchronization error of received time synchronization data packets. for The cumulative synchronization error, for Transmit time synchronization data packets to The resulting single-slot error; And / or, the electrocarbon information transmission model is constructed as follows: ， in, Indicates the first In the next iteration, the sensing device Transmit carbon information to the edge server connected to the smart fusion terminal The rate of transmission of electrical carbon information at that time; For equipment Task unloading decision variable, when the first iteration Choose to offload the carbon information to the server connected to the smart fusion terminal. To perform carbon dioxide calculations, then Conversely, ; And / or, the total time delay model for calculating the electrocarbon is constructed as follows: ， ， ， ， This indicates the total calculation delay for the carbon dioxide generator; This indicates the delay in unloading the carbon information. This indicates the calculation delay of the carbon information; For equipment The amount of collected carbon dioxide information data; For edge servers The number of computation cycles required to process each bit of carbon information; For edge servers For equipment The allocated computing resources; Edge servers for connecting smart converged terminals Total computing resources on board.

7. The method for synchronizing the calculation time of carbon dioxide according to claim 6, characterized in that, in, The optimization objectives P1 and P2 of the optimization problem are established as follows: ， In P1, Indicates the range of values ​​for the timing decision indicator variable; This indicates that only one exists in each iteration. For equipment Its timing, Indicates when and Time synchronization is not possible without topology connections; To constrain the transmission delay of time synchronization data packets; ， exist middle, This indicates the range of values ​​for the variable indicating the decision to unload the carbon information task. Indicates the device in each iteration The carbon information is offloaded to the edge server connected to the smart fusion terminal. To complete the carbon dioxide calculation; express and Without topology connections, task unloading is not possible; Indicates the server connected to the smart terminal A maximum of 100 can be simultaneously Each device performs carbon dioxide calculation; And / or, the calculation of the carbon information task offloading and computing resource allocation includes: Step 1: Obtain the sensing device For edge servers Preference value Based on preference values ​​by intelligent fusion terminals Sort all sensors in descending order to obtain a preference list for sensor devices. Among them, preference value The reciprocal of the total delay for the carbon dioxide is calculated as follows: ， Step 2: Obtain a decision through exchange matching calculation, including: initializing device task unloading decision traversal. Preference list Matching relationship utility function Then, the preference values ​​are calculated, and the preference list is updated. Secondly, a temporary matching relationship is established, with the intelligent fusion terminal serving as the device. Send a request to the server, when the edge server The number of received requests is less than At that time, the edge server Accept device requests, if the edge server The number of received requests is greater than Then accept the previous one. The system accepts one request, rejects requests from other devices, and updates the matching relationships. Further, construct exchange matching relationships based on the established temporary matching relationships. For equipment and Reallocate server computing resources and update utility functions. If, after swapping matching relationships, the utility functions of all devices are... If the value increases, the device update matching relationship will be as follows: and At the same time, update the task unloading decision variables and clear the swap matching relationship. Finally, when no exchange matching relationship exists, the matching terminates, completing the intelligent fusion terminal's decision-making for the unloading of carbon information tasks and resource allocation.

8. An intelligent fusion terminal implementing the time synchronization method for carbon dioxide calculation according to any one of claims 1-7, comprising an application layer, a physical layer, and a protocol interface layer, wherein, The application layer includes a topology-aware container for obtaining the topology information, a time synchronization container for completing the synchronization and timing, and an electric carbon computing container for obtaining the electric carbon information task offloading strategy and the electric carbon computing resource allocation scheme. The physical layer is used to support the operation of the intelligent fusion terminal and the application layer to realize its functions, and to connect with multiple edge servers; the protocol interface layer is used to realize information interaction and interaction information processing between the application layer and the physical layer.

Images