A method, device and medium for omnidirectional sensing fusion adaptive networking

Abstract

The application discloses a kind of general sense fusion adaptive networking method, equipment and medium, method includes: constructing general sense fusion queue model;Based on general sense fusion queue model, construct double time scale priority model;Based on general sense fusion queue model, construct general sense fusion transmission model;Based on general sense fusion transmission model and double time scale priority model, establish double time scale optimization target, and set constraint function;Through the IDQN algorithm of double time scale to double time scale optimization target for solving, construct double closed loop mechanism, carry out general sense fusion adaptive networking, through double time scale priority model, it can realize differentiated resource allocation: higher priority is given to key nodes, sudden change nodes and millisecond level control related nodes, while lower sampling frequency and lower access priority are used for general monitoring nodes, so as to preferentially guarantee timely discovery of key changes under limited communication and perception resources.

Description

Technical Field

[0001] This invention relates to the field of power cyber-physical systems technology, specifically to a sensor-integrated adaptive networking method, device, and medium. Background Technology

[0002] With the continuous advancement of the construction of new power systems, new elements such as distributed photovoltaic, wind power, energy storage systems, electric vehicles, adjustable loads, and virtual power plants are being connected to the distribution network on a large scale. The power grid operation mode is gradually changing from the traditional "source follows load" to a multi-entity collaborative interaction of "source, grid, load, and storage". At the same time, these new elements are characterized by dispersed access locations, variable operating states, and strong volatility and uncertainty, making the distribution network operation state more dynamic, complex, and coupled.

[0003] In the scenario of sensor fusion for hierarchical control of distribution networks, the system typically needs to make joint decisions across multiple dimensions, including terminal access authorization, sensing data acquisition frequency, communication bandwidth allocation, edge node collaboration, sensing task scheduling, and network relationship adjustment. Essentially, this is a multivariable, highly coupled, and strongly constrained dynamic optimization problem. Because different power grid terminals exhibit significant differences in service priorities, state sensitivity, and link resource occupancy, the configuration process of sensor resources is not only constrained by communication network capacity, latency, and reliability, but also needs to consider sensing accuracy, timeliness of state determination, and control response requirements.

[0004] Most existing methods employ static strategies with fixed-period updates, meaning that state assessments and resource adjustments are performed at a uniform pace within a preset time interval. This fails to fully consider the significant differences in communication resource allocation, sensing task updates, and power grid state evolution over time. For example, some parameter optimizations and threshold corrections are suitable for execution over longer time scales, while link switching, sampling frequency adjustments, and the reallocation of critical terminal resources often need to be completed rapidly over shorter time scales. If a single-time-scale strategy update mechanism is still used, it can easily lead to asynchrony between resource scheduling and state changes, resulting in delayed state change detection and untimely identification of critical anomalies, thereby affecting the accuracy and response speed of distribution network hierarchical control. Summary of the Invention

[0005] This invention proposes a sensor fusion adaptive networking method, device, and medium to solve the problems mentioned in the background.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A sensor fusion adaptive networking method of the present invention includes the following steps: S1. Construct a synesthetic fusion queue model; S2. Construct a dual-time-scale priority model based on the synesthetic fusion queue model; S3. Construct a synesthetic fusion transmission model based on the synesthetic fusion queue model; S4. Based on the synesthetic fusion transmission model and the dual-time-scale priority model, establish a dual-time-scale optimization objective and set constraint functions; S5. Solve the dual-time-scale optimization objective using the dual-time-scale IDQN algorithm, construct a dual-layer closed-loop mechanism, and perform sensor fusion adaptive networking.

[0007] Preferably, the sensor fusion queue model includes a set of power grid nodes, the amount of sensing data arriving at the power grid nodes, the local sensing cache queue of the power grid nodes, and the intensity of changes in the power grid nodes. The set of power grid nodes:

[0008] in, For the first Each power grid node corresponds to one or more sensing terminals, used to collect multi-source heterogeneous information of the power grid node, including voltage, current, frequency, phase angle, power, switch status, temperature, meteorological and environmental disturbances; The amount of sensing data arriving at the power grid node:

[0009] in, For nodes The set of perceived information types For the first The selection variable for whether class information is collected. This refers to the amount of data collected in a single sampling. For the corresponding sampling frequency, For discrete continuous time slots in the power cyber-physical system, The time slot length, ; The local sensing cache queue of the power grid node:

[0010] in, Indicates in time slot internal nodes The actual amount of data uploaded. In time slot internal nodes Locally aware cache queue; The intensity of change at the power grid node:

[0011] in, These represent the measured values ​​of node voltage, current, active power, and frequency, respectively. Indicates the trigger flags for discrete events such as switch position changes, protection actions, topology switching, and islanding formation. , , , and These correspond to the weighting coefficients of node voltage changes, current changes, active power changes, frequency changes, and discrete event triggering terms, respectively.

[0012] Preferably, the dual-timescale priority model includes a large-timescale priority model and a small-timescale priority model; The large-timescale priority model is as follows: On a large timescale, a regulation cycle is formed by multiple hourly slots and used to update the long-term priority of nodes, decision thresholds, and strategy parameters. The long-term priority of grid nodes is defined as follows:

[0013] in, For a large time scale. Indicates the electrical importance of the node. Indicates the average intensity of change over a large time scale. This indicates the frequency of historical alarms or abnormal events occurring at this node. Indicates the weight of business categories. , , , All weighting coefficients are non-negative. The small time-scale priority model is as follows: At a small time scale, the short-time priority of a power grid node is defined as follows:

[0014] in, For a small time scale, This is the maximum queue capacity. The remaining tolerable delay for state determination This is the maximum allowable decision delay for this type of service. Indicates the immediate urgency of business operations. These are the weighting coefficients; The overall priority of the constructed power grid nodes is as follows:

[0015] in, and This is a non-negative weighting coefficient used to merge the long-term and short-term priorities of nodes.

[0016] Preferably, the synergistic transmission model is as follows: The instantaneous transmission rate of a power grid node is expressed as:

[0017] in, For nodes In the time slot Indicator variables for authorized network access, For nodes In the time slot The allocated bandwidth, For nodes In the time slot Transmission power, For nodes In the time slot Channel gain, For nodes In the time slot The noise power spectral density; The effective data upload volume of a power grid node in a time slot is:

[0018] in, This indicates the total amount of data available for upload within the current time slot. This represents the maximum amount of data that the link can transmit within a time slot.

[0019] Preferably, the dual-timescale optimization objective includes an immediate optimization objective at a small timescale and a periodic utility objective at a large timescale; The immediate optimization objective at a small time scale is as follows:

[0020] in, For nodes In the time slot The change determines the success and profit. These are the non-negative weight coefficients for the change determination benefit item, resource utilization benefit item, state determination delay penalty item, and default penalty item, respectively. For nodes In the time slot Internal resource allocation efficiency or priority service benefits node In the time slot The comprehensive penalty for breach of contract caused by internal factors such as excessive latency, excessive bandwidth, incorrect access, or invalid high-frequency sampling. For nodes In the time slot State determination delay; The periodic utility objective over a large timescale is as follows:

[0021] in, For false alarm rate, The false negative rate, as well as These are the penalty weights for false positive rate and false negative rate.

[0022] Preferably, the constraint function includes:

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] in, The sampling frequency for power grid nodes. To represent nodes A set of selectable sampling frequencies To allocate bandwidth limits, For nodes Maximum transmit power limit For the set of key nodes, With a latency threshold in the millisecond range, This represents the lower limit of accuracy for determining node status. To observe the duration or the length of the statistical time window, For random variables The mathematical expectation, For nodes At any moment The accuracy of state determination. Access authorization constraints, Constraints are selected for the sampling frequency. Due to bandwidth resource constraints, For transmit power constraints, For millisecond-level state determination delay constraints, For queue stability constraints, This is a constraint on the accuracy of state determination.

[0030] Preferably, solving the dual-timescale optimization objective using the dual-timescale IDQN algorithm includes: On a large timescale, an upper-level intelligent agent is established, and the state space of the upper-level intelligent agent is defined as follows:

[0031] in, To average queue pressure over a period of time, The average intensity of change, To average the judgment delay, Average channel quality; The upper-level action space is defined as:

[0032] in, This is the state determination threshold vector. This is a set of priority weights for nodes of different categories. This is the update interval for the target network of the lower-layer IDQN; The large-scale reward function is defined as follows:

[0033] At a small time scale, a lower-level intelligent agent is established, and the state space of the lower-level intelligent agent is defined as follows:

[0034] The action space of the lower-level intelligent agent is defined as follows:

[0035] The definition of instantaneous reward on a small timescale is: .

[0036] Preferably, the IDQN algorithm employs a dual-network structure, including an evaluation network and a target network; The objective Q-value gives the form of the dual DQN objective: ) in, Discount factor; The loss function is:

[0037] The network parameters are updated using gradient descent.

[0038] The update interval is output by a large-scale agent, and the target network is updated after each hourly update interval. .

[0039] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0040] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0041] As can be seen from the above technical solution, the present invention provides a sensor fusion adaptive networking method. Compared with the prior art, the present invention has the following advantages: 1. This invention enables differentiated resource allocation through a dual-time-scale priority model: giving higher priority to key nodes, nodes experiencing sudden changes, and millisecond-level control-related nodes, while using lower sampling frequencies and lower access priorities for general monitoring nodes, thereby prioritizing the timely detection of "key changes" under limited communication and sensing resources.

[0042] 2. This invention adjusts terminal access authorization, sensing sampling frequency, and communication bandwidth in conjunction with node priority results, enabling critical nodes and nodes experiencing sudden changes to preferentially obtain higher sampling capabilities and transmission resources. The transmission rate of a node is jointly determined by its access status, allocated bandwidth, transmission power, and channel conditions, while the effective uploaded data volume is jointly determined by queue backlog and link transmission capacity. Through the above mechanism, a correlation is established between sampling frequency, bandwidth allocation, link transmission capacity, and state determination latency, enabling communication resource configuration to directly serve the rapid discovery and millisecond-level determination of state changes.

[0043] 3. The large-scale intelligent agent of this invention learns the threshold configuration, priority correction and network update rhythm under different operating stages through periodic interaction, so that fixed threshold, fixed priority and fixed network update cycle can adapt to the problem of dynamic evolution of power grid state; the lower-level intelligent agent outputs the optimal action according to the current state, so that key nodes and nodes with sudden changes can obtain high-frequency sampling and bandwidth guarantee first, so as to achieve millisecond-level state determination.

[0044] 4. This invention forms a two-layer closed-loop mechanism that adjusts parameters on a large time scale and makes decisions on a small time scale: the upper layer is responsible for slow variable adaptation, and the lower layer is responsible for fast variable millisecond-level response, thereby realizing sensor fusion adaptive networking for the rapid evolution of the state of the power information physical system. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a sensor fusion adaptive networking method according to the present invention.

[0046] Figure 2 This is a convergence curve of the state determination accuracy of the present invention.

[0047] Figure 3 This is a comparison chart of the average state determination delay under the method of the present invention and the traditional method.

[0048] Figure 4 This is a graph showing the changes in critical node availability and bandwidth utilization in this invention.

[0049] Figure 5 This is a comparison chart of the false alarm rate and false negative rate under the method of the present invention and the traditional method. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0051] like Figure 1 As shown, an adaptive networking method for sensor fusion in this embodiment includes the following steps: S1. Construct a synesthetic fusion queue model; S2. Construct a dual-time-scale priority model based on the synesthetic fusion queue model; S3. Construct a synesthetic fusion transmission model based on the synesthetic fusion queue model; S4. Based on the synesthetic fusion transmission model and the dual-time-scale priority model, establish a dual-time-scale optimization objective and set constraint functions; S5. Solve the dual-time-scale optimization objective using the dual-time-scale IDQN algorithm, construct a dual-layer closed-loop mechanism, and perform sensor fusion adaptive networking.

[0052] Furthermore, the sensor fusion queue model includes a set of power grid nodes, the amount of sensing data arriving at the power grid nodes, the local sensing cache queue of the power grid nodes, and the intensity of changes in the power grid nodes. Set of power grid nodes:

[0053] in, For the first Each power grid node corresponds to one or more sensing terminals, which are used to collect multi-source heterogeneous information of the power grid node, including voltage, current, frequency, phase angle, power, switch status, temperature, meteorological and environmental disturbances. The amount of sensing data arriving at power grid nodes:

[0054] in, For nodes The set of perceived information types For the first The selection variable for whether class information is collected. This refers to the amount of data collected in a single sampling. For the corresponding sampling frequency, For discrete continuous time slots in the power cyber-physical system, The time slot length, ; To reflect the waiting and transmission process of sensed data at the node and edge sides, a local sensed cache queue is established for each node to represent the backlog of data that has not yet been uploaded and judged. The local sensed cache queue of the power grid node is as follows:

[0055] in, Indicates in time slot internal nodes The actual amount of data uploaded. In time slot internal nodes Locally aware cache queue; To highlight the core issue of "change detection," this invention defines a node change intensity index. This characterizes the degree of abrupt change in the operating state of a node compared to the previous moment, representing the intensity of change in a power grid node.

[0056] in, These represent the measured values ​​of node voltage, current, active power, and frequency, respectively. Indicates the trigger flags for discrete events such as switch position changes, protection actions, topology switching, and islanding formation. , , , and These correspond to the weighting coefficients for node voltage changes, current changes, active power changes, frequency changes, and discrete event triggering terms, respectively. Each weighting coefficient is a non-negative real number. ≥0 (m=1,2,3,4,5); can be set as + + + + =1, in order to achieve normalized weighted fusion of various change characteristics.

[0057] Therefore, queue backlog not only reflects data congestion but is also coupled with the intensity of state changes, jointly determining the priority of subsequent resource allocation and state determination. Through this queue model, this invention integrates "whether a change occurs, how fast the change occurs, and whether the changed data is delivered in a timely manner" into a quantifiable framework.

[0058] Considering the significant differences in importance, business type, and rate of change among different nodes in the power system, this invention constructs a dual-time-scale priority model.

[0059] Furthermore, the dual-timescale priority model includes a large-timescale priority model and a small-timescale priority model; The large-scale priority model is as follows: On a large timescale, a regulation cycle is formed by multiple hourly slots and used to update the long-term priority of nodes, decision thresholds, and strategy parameters. The long-term priority of grid nodes is defined as follows:

[0060] in, For a large time scale. This indicates the electrical importance of nodes, reflecting their topological criticality within the system, such as substation outlets, tie points, key distributed power source access points, and important load nodes. Indicates the average intensity of change over a large time scale. This indicates the frequency of historical alarms or abnormal events occurring at this node. This indicates the weight of the service category, used to distinguish different service levels such as protection information, control variables, operational status variables, and environmental auxiliary variables. , , , The weight coefficients are all non-negative and satisfy the normalization constraint, that is, each weight coefficient belongs to the interval [0,1] and their sum is 1; the weights of the four indicators can be 0.35, 0.30, 0.20 and 0.15 respectively; Large timescales primarily address questions such as "who is more important," "which nodes deserve long-term attention," and "how to adjust the judgment thresholds based on changes in historical status." The small-time-scale priority model is as follows: At a small time scale, the short-time priority of a power grid node is defined as follows:

[0061] in, For a small time scale, This is the maximum queue capacity. The remaining tolerable delay for state determination This is the maximum allowable decision delay for this type of service. Indicates the immediate urgency of business operations. The weights are set as coefficients and satisfy the normalization constraint, meaning that each weight coefficient belongs to the interval [0,1] and their sum is 1; the weight related to queue occupancy can be 0.40, and the weight related to latency urgency can be 0.60. The overall priority of the constructed power grid nodes is as follows:

[0062] in, and This is a non-negative weighting coefficient used to merge the long-term and short-term priorities of nodes.

[0063] In summary, through this dual-time-scale priority model, the present invention can achieve differentiated resource allocation: giving higher priority to key nodes, nodes with sudden changes, and millisecond-level control-related nodes, while using lower sampling frequency and lower access priority for general monitoring nodes, thereby prioritizing the timely detection of "key changes" under limited communication and sensing resources.

[0064] In the context of sensor fusion, sensing is not performed in isolation, but is directly coupled with communication link capacity and bandwidth scheduling. Therefore, this invention establishes a joint communication-sensing transmission model.

[0065] Furthermore, the sensory fusion transmission model is as follows: The instantaneous transmission rate of a power grid node is expressed as:

[0066] in, For nodes In the time slot Indicator variables for authorized network access, For nodes In the time slot The allocated bandwidth, For nodes In the time slot Transmission power, For nodes In the time slot Channel gain, For nodes In the time slot The noise power spectral density; The effective data upload volume of a power grid node in a time slot is:

[0067] in, This indicates the total amount of data available for upload within the current time slot. This represents the maximum amount of data that the link can transmit within a time slot.

[0068] In a sensor fusion scenario, the sensing process is coupled with communication link capacity and bandwidth scheduling. Based on node priority, terminal access authorization, sensing sampling frequency, and communication bandwidth are jointly adjusted to prioritize critical nodes and nodes experiencing sudden changes in access to higher sampling capabilities and transmission resources. A node's transmission rate is determined by its access status, allocated bandwidth, transmit power, and channel conditions, while the effective uploaded data volume is determined by queue backlog and link transmission capacity. Through this mechanism, a correlation is established between sampling frequency, bandwidth allocation, link transmission capacity, and state determination latency, enabling communication resource allocation to directly serve the rapid detection and millisecond-level determination of state changes.

[0069] Furthermore, the dual-timescale optimization objective includes an immediate optimization objective at a small timescale and a periodic utility objective at a large timescale. The immediate optimization objective at a small time scale is as follows:

[0070] in, For nodes In the time slot The change determines the success and profit. These are the non-negative weight coefficients for the change determination benefit item, resource utilization benefit item, state determination delay penalty item, and default penalty item, respectively, satisfying... ≥0、 ≥0、 ≥0、 ≥0; in one embodiment, it can be further set to + + + =1, so as to achieve normalized weighting among different objective terms. For nodes In the time slot Internal resource allocation efficiency or priority service benefits node In the time slot The comprehensive penalty for breach of contract caused by internal factors such as excessive latency, excessive bandwidth, incorrect access, or invalid high-frequency sampling. For nodes In the time slot State determination delay; The periodic utility objective over a large timescale is as follows:

[0071] in, For false alarm rate, The false negative rate, as well as These are the penalty weights for false positive rate and false negative rate.

[0072] Furthermore, the constraint functions include:

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] in, The sampling frequency for power grid nodes. To represent nodes A set of selectable sampling frequencies To allocate bandwidth limits, For nodes Maximum transmit power limit For the set of key nodes, With a latency threshold in the millisecond range, This represents the lower limit of accuracy for determining node status. To observe the duration or the length of the statistical time window, For random variables The mathematical expectation, For nodes At any moment The accuracy of state determination. Access authorization constraints, Constraints are selected for the sampling frequency. Due to bandwidth resource constraints, For transmit power constraints, For millisecond-level state determination delay constraints, For queue stability constraints, This is a constraint on the accuracy of state determination.

[0080] Through the above objective function and constraint system, this invention integrates "whether the discovery of changes is timely, accurate, and worthwhile to expend more resources" into the same optimization framework.

[0081] To address the aforementioned problems, which are characterized by high-dimensional state space, strong dynamics, and nonlinear coupling, and are not suitable for direct solutions using traditional convex optimization, this invention employs a dual-time-scale improved deep Q-network for online decision-making.

[0082] Furthermore, the dual-timescale optimization objective is solved using the dual-timescale IDQN algorithm, including: On a large timescale, an upper-level intelligent agent is established, and the state space of the upper-level intelligent agent is defined as follows:

[0083] in, To average queue pressure over a period of time, The average intensity of change, To average the judgment delay, Average channel quality; The upper-level action space is defined as:

[0084] in, This is the state determination threshold vector. This is a set of priority weights for nodes of different categories. This is the update interval for the target network of the lower-layer IDQN; The large-scale reward function is defined as follows:

[0085] At a small time scale, a lower-level intelligent agent is established, and the state space of the lower-level intelligent agent is defined as follows:

[0086] Large-scale intelligent agents learn threshold configurations, priority adjustments, and network update rhythms at different operational stages through periodic interactions, thereby solving the problem that fixed thresholds, fixed priorities, and fixed network update cycles cannot adapt to the dynamic evolution of the power grid state.

[0087] The action space of the lower-level intelligent agent is defined as follows:

[0088] The definition of instantaneous reward on a small timescale is:

[0089] The lower-level intelligent agent outputs the optimal action based on the current state, so that key nodes and nodes with sudden changes can obtain high-frequency sampling and bandwidth guarantee first, so as to achieve millisecond-level state determination.

[0090] Furthermore, the IDQN algorithm employs a dual-network structure, including an evaluation network and a target network; The objective Q-value gives the form of the dual DQN objective: ) in, This is a discount factor used to characterize the contribution of future returns to the current target Q value; The larger the value, the more the agent values ​​long-term cumulative benefits. The smaller the value, the more the agent focuses on the immediate benefits of the current time slot. ∈[0.90,0.99], when emphasizing long-term networking performance and the effect of continuous state determination, 0.95 or 0.99 can be taken; The loss function is:

[0091] The network parameters are updated using gradient descent.

[0092] The update interval is output by a large-scale agent, and the target network is updated after each hourly update interval.

[0093] Furthermore, to improve the learning efficiency of samples of sudden state changes, this invention introduces a priority experience replay mechanism based on the intensity of change and the degree of delay default. Samples with high change intensity and high risk of delay exceeding limits are given higher sampling probabilities, enabling the network to learn the decision-making rules of how to prioritize the discovery of key changes more quickly.

[0094] Thus, this invention forms a two-layer closed-loop mechanism of "adjusting parameters on a large time scale and making decisions on a small time scale": the upper layer is responsible for slow variable adaptation, and the lower layer is responsible for fast variable millisecond-level response, thereby realizing sensor fusion adaptive networking for the rapid evolution of the state of the power information physical system.

[0095] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0096] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0097] In practical applications, experimental tests were conducted on a constructed distribution network hierarchical control simulation platform. The simulation platform includes 30 grid nodes, 120 sensing terminals, 4 edge coordination units, and limited wireless communication links, comprehensively considering typical operating factors such as load fluctuations, distributed power generation output changes, limited communication bandwidth, and random disturbance events.

[0098] In the experiments, the dual-timescale IDQN method proposed in this invention was compared with the fixed-period DQN method and the static strategy method. The evaluation metrics mainly included state determination accuracy, average state determination latency, critical node guarantee rate, bandwidth utilization, false alarm rate, and missed alarm rate. Each method was repeatedly run under the same network topology, the same service disturbance conditions, and the same bandwidth constraints. The results were averaged from multiple independent experiments to ensure consistency and comparability of the comparison results.

[0099] The experimental platform and parameter settings are as follows: Table 1 Simulation Parameter Settings Parameter name symbol Setting value illustrate Number of power grid nodes N 30 Total number of distribution network hierarchical control nodes Number of sensing terminals M 120 Total number of sensing terminals deployed at each node Number of edge collaborative units E 4 Responsible for local coordination and resource scheduling Total communication bandwidth B 20MHz Available wireless communication bandwidth of the system Experience replay capacity D 5000 IDQN Experience Replay Cache Capacity Discount factor γ 0.95 Balancing immediate and long-term returns First, we examine the convergence of state determination accuracy of different methods during the training process. Figure 2 It can be seen that the proposed dual-timescale IDQN method can quickly improve the accuracy of state determination in the early stage of training and tend to stabilize after about 150 rounds, eventually stabilizing at a high level. This shows that the dual-layer mechanism of adjusting parameters on a large timescale and making decisions on a small timescale can more effectively adapt to the evolution of system state and improve the recognition accuracy of key state changes.

[0100] Figure 3 The average state determination latency of different methods is compared. It can be seen that the static strategy method has the highest average state determination latency. Although the fixed-period DQN method has some improvement, it is still difficult to balance fast determination and resource coordination. The proposed dual-timescale IDQN method further reduces the average state determination latency and can better meet the millisecond-level state determination requirements.

[0101] Figure 4 The results show the changes in critical node guarantee rate and bandwidth utilization rate over the scheduling cycle. As the scheduling cycle progresses, both the critical node guarantee rate and bandwidth utilization rate of the proposed method show a stable increasing trend, indicating that the method can prioritize the guarantee of high-importance nodes and nodes with sudden changes under limited communication resources, while improving the overall bandwidth allocation efficiency.

[0102] To further evaluate the reliability of the proposed method, the false positive and false negative rates of different methods on the test set were statistically analyzed; Figure 5 It can be seen that the proposed dual-timescale IDQN method outperforms the comparative methods in both metrics, indicating that it can not only improve the accuracy and timeliness of state determination, but also effectively suppress false alarms and false negatives, and enhance the stability and robustness of the system in operation.

[0103] The experimental results show that the dual-timescale improved deep reinforcement learning algorithm proposed in this invention can achieve high state determination accuracy, low average determination latency, high critical node protection rate, and superior bandwidth utilization in sensor fusion scenarios, while exhibiting better overall performance in terms of false alarm rate and false negative rate. This indicates that the method is suitable for millisecond-level state determination and adaptive networking scenarios in power cyber-physical systems, and can provide effective technical support for hierarchical control of distribution networks.

[0104] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk), etc.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0106] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0107] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sensor fusion adaptive networking method, characterized in that, Includes the following steps: S1. Construct a synesthetic fusion queue model; S2. Based on the sensor fusion queue model, a dual time scale priority model is constructed. The sensor fusion queue model includes a set of power grid nodes, the amount of sensing data arriving at the power grid nodes, the local sensing cache queue of the power grid nodes, and the intensity of changes in the power grid nodes. The dual-timescale priority model includes a large-timescale priority model and a small-timescale priority model. The large-timescale priority model is as follows: On a large timescale, a regulation cycle is formed by multiple hourly slots and used to update the long-term priority of nodes, decision thresholds, and strategy parameters. The long-term priority of grid nodes is defined as follows: in, For a large time scale. Indicates the electrical importance of the node. Indicates the average intensity of change over a large time scale. This indicates the frequency of historical alarms or abnormal events occurring at this node. Indicates the weight of business categories. , , , All weighting coefficients are non-negative. The small time-scale priority model is as follows: At a small time scale, the short-time priority of a power grid node is defined as follows: in, For a small time scale, This is the maximum queue capacity. The remaining tolerable delay for state determination This is the maximum allowable decision delay for this type of service. Indicates the immediate urgency of business operations. These are the weighting coefficients; The overall priority of the constructed power grid nodes is as follows: in, and These are non-negative weighting coefficients used to combine the long-term and short-term priorities of nodes; S3. Construct a synesthetic fusion transmission model based on the synesthetic fusion queue model; S4. Based on the synesthetic fusion transmission model and the dual-time-scale priority model, establish a dual-time-scale optimization objective and set constraint functions; S5. Solve the dual-time-scale optimization objective using the dual-time-scale IDQN algorithm, construct a dual-layer closed-loop mechanism, and perform sensor fusion adaptive networking.

2. The sensor fusion adaptive networking method according to claim 1, characterized in that: The set of power grid nodes: in, For the first Each power grid node corresponds to one or more sensing terminals, used to collect multi-source heterogeneous information of the power grid node, including voltage, current, frequency, phase angle, power, switch status, temperature, meteorological and environmental disturbances; The amount of sensing data arriving at the power grid node: in, For nodes The set of perceived information types For the first The selection variable for whether class information is collected. This refers to the amount of data collected in a single sampling. For the corresponding sampling frequency, For discrete continuous time slots in the power cyber-physical system, The time slot length, ; The local sensing cache queue of the power grid node: in, Indicates in time slot internal nodes The actual amount of data uploaded. In time slot internal nodes Locally aware cache queue; The intensity of change at the power grid node: in, These represent the measured values ​​of node voltage, current, active power, and frequency, respectively. Indicates the trigger flags for discrete events such as switch position changes, protection actions, topology switching, and islanding formation. , , , and These correspond to the weighting coefficients of node voltage changes, current changes, active power changes, frequency changes, and discrete event triggering terms, respectively.

3. The sensor fusion adaptive networking method according to claim 2, characterized in that: The synergistic transmission model is as follows: The instantaneous transmission rate of a power grid node is expressed as: in, For nodes In the time slot Indicator variables for authorized network access, For nodes In the time slot The allocated bandwidth, For nodes In the time slot Transmission power, For nodes In the time slot Channel gain, For nodes In the time slot The noise power spectral density; The effective data upload volume of a power grid node in a time slot is: in, This indicates the total amount of data available for upload within the current time slot. This represents the maximum amount of data that the link can transmit within a time slot.

4. The sensor fusion adaptive networking method according to claim 3, characterized in that: The dual-timescale optimization objective includes an immediate optimization objective at a small timescale and a periodic utility objective at a large timescale. The immediate optimization objective at a small time scale is as follows: in, For nodes In the time slot The change determines the success and profit. These are the non-negative weight coefficients for the change determination benefit item, resource utilization benefit item, state determination delay penalty item, and default penalty item, respectively. For nodes In the time slot The efficiency of resource allocation or the benefits of priority services within the region. node In the time slot The comprehensive penalty for breach of contract caused by internal factors such as excessive latency, excessive bandwidth, incorrect access, or invalid high-frequency sampling. For nodes In the time slot State determination delay; The periodic utility objective over a large timescale is as follows: in, For false alarm rate, The false negative rate, as well as These are the penalty weights for false positive rate and false negative rate.

5. The sensor fusion adaptive networking method according to claim 4, characterized in that: The constraint functions include: in, The sampling frequency for power grid nodes. To represent nodes A set of selectable sampling frequencies To allocate bandwidth limits, For nodes Maximum transmit power limit For the set of key nodes, With a latency threshold in the millisecond range, This represents the lower limit of accuracy for determining node status. To observe the duration or the length of the statistical time window, For random variables The mathematical expectation, For nodes At any moment The accuracy of state determination. Access authorization constraints, Constraints are selected for the sampling frequency. Due to bandwidth resource constraints, For transmit power constraints, For millisecond-level state determination delay constraints, For queue stability constraints, This is a constraint on the accuracy of state determination.

6. The sensor fusion adaptive networking method according to claim 5, characterized in that: The solution of the dual-timescale optimization objective using the dual-timescale IDQN algorithm includes: On a large timescale, an upper-level intelligent agent is established, and the state space of the upper-level intelligent agent is defined as follows: in, To average queue pressure over a period of time, The average intensity of change, To average the judgment delay, Average channel quality; The upper-level action space is defined as: in, This is the state determination threshold vector. This is a set of priority weights for nodes of different categories. This is the update interval for the target network of the lower-layer IDQN; The large-scale reward function is defined as follows: At a small time scale, a lower-level intelligent agent is established, and the state space of the lower-level intelligent agent is defined as follows: The action space of the lower-level intelligent agent is defined as follows: The definition of instantaneous reward on a small timescale is: 。 7. The sensor fusion adaptive networking method according to claim 6, characterized in that: The IDQN algorithm employs a dual-network structure, including an evaluation network and a target network; The objective Q-value gives the form of the dual DQN objective: ) in, Discount factor; The loss function is: The network parameters are updated using gradient descent. The update interval is output by a large-scale agent, and the target network is updated after each hourly update interval. 。 8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images