Power distribution network fault self-healing control method based on multi-agent cooperation

Abstract

The application discloses a power distribution network fault self-healing control method based on multi-agent cooperation and relates to the technical field of power distribution networks. The method comprises the following steps: acquiring steady-state power data of multiple communication nodes in a power distribution network before fault tripping; calculating expected power data based on the steady-state power data and power outage duration; performing message iteration processing on adjacent communication nodes based on the expected power data and a Gaussian belief propagation algorithm to obtain expected load evaluation values and data loss degrees of the communication nodes; calculating a risk degree value of cable overload of the communication nodes based on the expected load evaluation values and the data loss degrees; and when the maximum value of the risk degree value exceeds a preset safety threshold, controlling a target node to disconnect a local load access side breaker and controlling a tie switch to close after isolating an out-of-limit load. The application achieves the technical effect of avoiding the situation that blind closing easily causes main cable overload and overstep tripping.

Description

Technical Field

[0001] This application relates to the field of power distribution network technology, specifically to a power distribution network fault self-healing control method based on multi-agent cooperation. Background Technology

[0002] After a short-circuit fault occurs in the distribution network and the faulty section is disconnected, the system usually needs to close the backup tie switch on the main cable so that the backup line can restore power to the non-faulty area in order to achieve self-healing reconfiguration of the network.

[0003] However, during actual power outages, inductive loads gradually cool down, and the instant power is restored, a surge current (load bounce) far exceeding the steady-state current can be generated. Existing self-healing control schemes for distribution networks typically only collect and directly accumulate the steady-state active power before the fault occurs, using this as the basis for assessing the restoration capacity. However, under extreme weather or disaster conditions, the underlying communication environment of the distribution network is often unstable. Once local node data is missing, calculations often stall. If the missing data is skipped, the load will be overlooked, resulting in the expected capacity being far less than the actual closing impact capacity. Blindly closing the circuit can easily cause overload of the main cables and cascading tripping. Summary of the Invention

[0004] To address the technical problem in related technologies where the steady-state active power collected before a fault occurs is directly accumulated and used as the basis for evaluating the capacitance, blindly closing the circuit breaker can easily lead to overload of the main cable and cascading tripping, this application provides a self-healing control method for distribution network faults based on multi-agent cooperation.

[0005] The specific technical solution adopted is as follows:

[0006] The system acquires steady-state power data of multiple communication nodes in the distribution network before a fault trip. Based on the steady-state power data and the duration of the power outage, it calculates the expected power data of each communication node at the moment of power restoration. The communication nodes interact with each other through an intelligent agent.

[0007] Based on the expected power data and the Gaussian belief propagation algorithm, message iterative processing is performed on adjacent communication nodes to obtain the expected load assessment value and data missingness of the communication nodes.

[0008] Based on the expected load assessment value and the data missingness, the risk level of cable overload at the communication node is calculated.

[0009] When the maximum value of the risk level exceeds the preset safety threshold, the target node is controlled to disconnect the local load access side circuit breaker, and the tie switch is controlled to close after isolating the over-limit load. The target node is the node corresponding to the maximum value of the risk level.

[0010] In one possible implementation of this application, based on steady-state power data and the duration of the power outage, the expected power data of each communication node at the moment of power restoration is calculated, including:

[0011] Determine the power restoration surge coefficient and the power outage cooling coefficient;

[0012] Based on the power restoration surge coefficient, power outage cooling coefficient, and power outage duration, the surge multiple of the power data relative to the steady-state power data at the moment of power restoration is calculated;

[0013] Based on the surge multiple and steady-state power data, the expected power data of each communication node at the moment of power restoration is calculated.

[0014] In one possible implementation of this application, based on expected power data and a Gaussian belief propagation algorithm, message iteration processing is performed on adjacent communication nodes to obtain the expected load assessment value and data missingness of the communication nodes, including:

[0015] For any communication node, the expected power data is used as the initial local observation mean of the current node in the Gaussian belief propagation algorithm;

[0016] The local observation variance is calculated based on the square of the initial local observation mean.

[0017] In any iteration of the Gaussian belief propagation algorithm, the mean and variance of messages sent by each neighboring node are received;

[0018] The initial local observation mean, local observation variance, message mean, and message variance are combined to obtain the expected load assessment value and data missingness.

[0019] In one possible implementation of this application, the initial local observation mean, local observation variance, message mean, and message variance are combined to obtain the expected load assessment value and data missingness, including:

[0020] The local observation variance of the current node is combined with the message variance of all neighboring nodes in the previous round to obtain the comprehensive accuracy parameter of the current iteration round.

[0021] Based on the comprehensive accuracy parameter, the initial local observation mean, the local observation variance, and the weighted sum of the message mean and message variance of all adjacent nodes, the expected load assessment value for the current iteration round is calculated.

[0022] The data missingness of the current iteration is determined based on the reciprocal of the comprehensive accuracy parameter.

[0023] The loop terminates when the number of iterations reaches the preset maximum number of iterations, and the final expected load assessment value and data missingness are output.

[0024] In one possible implementation of this application, after receiving the mean and variance of messages sent by each neighboring node, the method further includes:

[0025] If no data is received from the target neighbor node after the preset waiting period, it is determined that a communication packet loss has occurred, and the message variance corresponding to the target neighbor node is forcibly assigned to a preset maximum constant, and its message mean is forcibly assigned to zero.

[0026] In one possible implementation of this application, a risk level value for cable overload at a communication node is calculated based on the expected load assessment value and the data missingness, including:

[0027] The expected load assessment value and the data missingness are weighted and summed to obtain the risk level of cable overload at the communication node.

[0028] In one possible implementation of this application, before controlling the target node to disconnect the local load access side circuit breaker when the maximum value of the risk level exceeds a preset safety threshold, the method further includes:

[0029] When the maximum value of the risk level does not exceed the preset safety threshold, the control switch will close.

[0030] In one possible implementation of this application, controlling the target node to disconnect the local load access side circuit breaker includes:

[0031] A branch disconnection command is sent to the target node via the tie switch. After receiving the branch disconnection command, the target node controls the local load access side circuit breaker to disconnect and execute the disconnection operation.

[0032] In one possible implementation of this application, after controlling the target node to disconnect the local load access side circuit breaker, the method further includes:

[0033] The control target node maintains the circuit breakers on both sides of the main line in a closed and connected state to ensure the smooth transmission of main line power to downstream healthy nodes.

[0034] In one possible implementation of this application, acquiring steady-state power data of multiple communication nodes in a distribution network before a fault trip includes:

[0035] Acquire active power and reactive power data of multiple communication nodes in the distribution network within a preset steady-state time period before fault tripping;

[0036] The active power data and reactive power data are integrated and processed to obtain steady-state power data.

[0037] This application has, but is not limited to, the following technical effects:

[0038] In this application, a Gaussian belief propagation algorithm is used for iterative interaction and to output the data missingness. This feature maintains node inference calculations even when data packets are lost in the communication network, avoiding computational stagnation caused by waiting for missing data in traditional distributed summation algorithms. The cable overload risk level is calculated by combining the expected load assessment value and the data missingness. Then, based on the maximum risk level exceeding the limit, the target node circuit breaker is controlled to open and the tie switch is closed. This allows for targeted load shedding based on the risk level value, reducing the control center's dependence on complete global data and preventing blind closing that could easily lead to trunk cable overload and cascading tripping. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the first embodiment of the self-healing control method for distribution network faults based on multi-agent cooperation in this application.

[0040] Figure 2 This is a schematic diagram of the equipment structure involved in the self-healing control method for power distribution network faults based on multi-agent cooperation in this application. Detailed Implementation

[0041] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0042] This application provides a distribution network fault self-healing control method based on multi-agent cooperation. In the first embodiment of the distribution network fault self-healing control method based on multi-agent cooperation in this application, referring to... Figure 1 The methods include:

[0043] Step S10: Obtain steady-state power data of multiple communication nodes in the distribution network before the fault trip. Based on the steady-state power data and the duration of the power outage, calculate the expected power data of each communication node at the moment of power restoration. The communication nodes interact with each other through an intelligent agent.

[0044] As an example, the multi-agent cooperative distribution network fault self-healing control method can also be applied to a multi-agent cooperative distribution network fault self-healing control system, in which the implementing agents include two types:

[0045] 1. Communication Node (hereinafter referred to as Node): The intelligent agent in the power distribution network. It is responsible for recording the power outage time, calculating the expected power data, establishing peer-to-peer communication links, executing the Gaussian belief propagation algorithm, calculating the risk level of cable overload, comparing and transmitting extreme values ​​back to the upstream unicast, and receiving instructions to disconnect the circuit breaker on the local load access side.

[0046] 2. Interconnection switch: The local controller on the backup power supply side of the main line is responsible for flood broadcast power restoration detection frames, receiving aggregated line overload assessment data packets, comparing the system's maximum cable overload assessment value with the system safety boundary constant, and executing closing actions or issuing directional branch disconnection commands.

[0047] As an example, steady-state power data is the sum of active and reactive power data during the steady-state period preceding the fault trip.

[0048] The method for obtaining steady-state power data is as follows:

[0049] Acquire active power and reactive power data of multiple communication nodes in the distribution network within a preset steady-state time period before fault tripping;

[0050] The active power data and reactive power data are integrated and processed to obtain steady-state power data.

[0051] As an example, when a short-circuit fault occurs in the distribution network, a node captures the tripping signal of the local circuit breaker. The node records the specific moment when the circuit breaker trips as the circuit breaker's start time. Regarding the start time of the circuit breaker's operation Extract the average active power data within a preset steady-state time period (e.g., 1 minute before tripping) before the tripping action (average active power). The average value of reactive power data (average reactive power) Active power data represents the electrical power actually consumed by the equipment and converted into other useful energy (such as mechanical energy, thermal energy, and light energy). Reactive power data represents the electrical power that is not actually consumed but is only used to exchange energy between the power source and inductive / capacitive electrical equipment (such as motor coils and transformers) to establish an alternating magnetic or electric field.

[0052] As an example, steady-state power data The calculation method is as follows:

[0053] in: This represents the average active power of the node during the preset steady-state period before tripping. This represents the average reactive power of the node during the preset steady-state period before tripping. By taking the square root of the sum of the squares of active and reactive power, the system can accurately reconstruct the true overall load capacity before tripping. Simultaneously, the node triggers an internal high-precision timer. This high-precision timer starts from the moment the circuit breaker operates. The power outage duration begins to accumulate continuously from zero, marking the node entering a power outage cooling state.

[0054] The step S10, which calculates the expected power data of each communication node at the moment of power restoration based on steady-state power data and the duration of the power outage, includes:

[0055] Determine the power restoration surge coefficient and the power outage cooling coefficient.

[0056] As an example, during the deployment or daily operation of distribution network equipment, the system pre-enters inherent electrical attribute parameters to each node, providing basic data for subsequent load rebound calculations. First, the system obtains the device type data of the locally connected loads at each node. Specifically, the parameters are obtained by consulting the factory-rated starting current multiple on the nameplate of the connected equipment, or by extracting current surge curves from historical power outage logs and performing offline data fitting. Based on the obtained data, the system stores the power rebound surge coefficient in the node's non-volatile memory. With power outage cooling coefficient .

[0057] It should be noted that the sudden increase coefficient of complex current This is a dimensionless constant used to quantify the degree of current surge caused by a power outage. Specific parameter configuration examples are as follows: For nodes with purely lighting loads, there is no back electromotive force within the equipment, and the system will... The value can be configured as any positive number between 0 and 0.2. For nodes that are equipped with inductive loads such as heavy industrial motors, the system will... The value can be configured from 4.0 to 7.0.

[0058] Power outage cooling coefficient Used to characterize the rate at which the rotational speed or internal temperature of equipment decreases over time due to a power outage. The entered power outage cooling coefficient... The time unit is configured to be the reciprocal of the time statistics unit of the node's local high-precision timer. For example, if the local timer uses seconds ( If the sum is accumulated in units of ), then The unit is configured as the negative first power of seconds ( Simultaneously, at the interconnection switch, the system reads the hardware specifications of the local power supply backbone cable and enters the cable's system safety boundary constant. Each node reads the local branch cable specifications or the trunk capacity allocation quota and enters the rated thermal limit capacity locally. In a specific parameter configuration embodiment, the system safety boundary constant Set to a fixed value of 1.0, its physical meaning represents the full-load critical point when the ratio of the evaluated capacity to the cable's rated thermal limit capacity reaches 100%.

[0059] Based on the power restoration surge coefficient, the power outage cooling coefficient, and the power outage duration, the surge multiple of the power data relative to the steady-state power data at the moment of power restoration is calculated.

[0060] Based on the surge multiple and steady-state power data, the expected power data of each communication node at the moment of power restoration is calculated.

[0061] As an example, after the faulty section is isolated by the circuit breaker, the tie switch floods a power restoration detection frame to the remaining outage network. Each node in the outage cooling state receives this power restoration detection frame. The node extracts the current cumulative value of its local high-precision timer at the moment the power restoration detection frame is received and marks this cumulative value as the outage duration. .

[0062] Specifically, the system continuously monitors the accumulated time of the high-precision timer and executes timeout protection logic: when the accumulated time of the high-precision timer reaches the preset maximum cooling threshold (e.g., configured to 24 hours), and the node still has not received a power restoration detection frame, the system triggers the timeout handling mechanism. The node no longer waits but directly records the power outage duration. This value is assigned to the maximum cooling threshold. This setting is used to prevent numerical overflow of the underlying timer and to provide a conservative estimate of the fault-tolerant time boundary for the system.

[0063] As an example, for the i-th node, the expected power data The calculation formula is expressed as:

[0064]

[0065] in: Indicates the coefficient of sudden increase in complex current. Indicates the cooling coefficient during power outages. Represents a decay term based on the natural logarithm; a single term. Indicates the degree of completion of internal cooling of electrical equipment after power failure; single item This indicates the dynamic increase ratio after inductive characteristics of the device are superimposed; combined term This indicates the sudden increase in total capacity relative to steady-state power data at the moment of power restoration; expected power data. This represents the actual line load capacity demand expected to occur at the moment the current node is powered on again.

[0066] The above formula shows that as the duration of the power outage increases... The increase of the exponential term The value gradually decays and approaches zero, resulting in a decrease in the cooling completion rate. The calculated result continuously increases and approaches 1. This objectively reflects the physical law that the longer the power outage and the more thorough the equipment cooling, the higher the proportion of capacity surge caused by restarting.

[0067] Step S20: Based on the expected power data and the Gaussian belief propagation algorithm, perform message iteration processing on adjacent communication nodes to obtain the expected load assessment value and data missingness of the communication nodes.

[0068] As an example, the expected load assessment value represents the power assessment value derived by the Gaussian belief propagation algorithm over multiple iterations, which is the baseline load base for the healthy network to reach consensus. The data missingness represents the degree of uncertainty of the data derived by the Gaussian belief propagation algorithm over multiple iterations. When data loss occurs, the data missingness will increase significantly.

[0069] As an example, the Gaussian belief propagation algorithm is used to replace conventional pipelined summation calculations in existing technologies. It maintains iteration without deadlock when node communication is interrupted, separating the output load projection baseline from the quantified value of packet loss scale. Communication packet loss events are mapped to abrupt changes in the algorithm's observation variance. Utilizing the mathematical property that packet loss injection causes precision collapse due to preset extreme values, the algorithm is forced to output a drastically expanded missing variance, preventing overload caused by omissions. Here, for any given node, adjacent communication nodes refer to all nodes directly physically connected to the current node.

[0070] Among them, step S20 of the distribution network fault self-healing control based on multi-agent cooperation also includes steps S21 to S24:

[0071] Step S21: For any communication node, use the expected power data as the initial local observation mean of the current node in the Gaussian belief propagation algorithm.

[0072] As an example, since the Gaussian belief propagation algorithm requires explicit initial means and variances to start, the system cannot automatically initiate iterative computation without defined initial conditions. Furthermore, nodes... Obtain expected power data ,node Register it as the initial local observation mean of the Gaussian belief propagation algorithm. .

[0073] Step S22: Calculate the local observation variance based on the square of the initial local observation mean.

[0074] As an example, at the same time, nodes Set local observation variance To prevent the algorithm from falling into numerical deadlock due to an excessively small variance constant in subsequent calculations, the system dynamically generates an initial variance value based on the initial load. Node Set local observation variance This is a constant multiple of the square of the expected power output upon initial power restoration. The specific calculation formula is as follows:

[0075]

[0076] in: Represents a node The expected power data, which is the initial local observation mean, This represents the preset local sensor observation uncertainty coefficient; The value is a very small positive constant (e.g., 0.001) used to prevent overflow in precision reciprocal calculations when the node is unloaded. For the preset uncertainty coefficient... The specific configuration implementation is as follows: the system, based on the factory accuracy error of the local electrical acquisition sensor at the node, preferably sets its value between 0.01 and 0.05. This calculation method shows that the larger the initial load of the node, the more proportionally its initial observation variance is amplified. This ensures that even under extreme single-node isolated operating conditions, the algorithm can still output a sufficiently large amount of penalized variance. As an example, for the node... Send to each adjacent node The initial message, node Its initial transit variance Set to a very large constant (as per the configuration example, set to) This maximum constant represents the state of the surrounding neighboring nodes being completely unknown at the start of the iteration.

[0077] Step S23: In any iteration of the Gaussian belief propagation algorithm, receive the mean and variance of messages sent by each neighboring node.

[0078] As an example, this step improves the traditional direct numerical summation into an iterative deduction based on probability distribution. After performing this operation, the system does not rely on complete neighbor node data, but instead utilizes the mathematical characteristic that the output variance of the Gaussian belief propagation algorithm increases sharply when communication packet loss occurs, thereby reducing the impact of packet loss on the statistical expected capacity.

[0079] The system sets the current iteration round as . (Initially, ), and set the maximum number of iterations constant to be (As a configuration example, the following is set) ). In the At the beginning of the next iteration, the node Attempt to receive each neighboring node from the set of neighboring communication nodes Average number of messages sent and message variance , where k-1 represents the previous iteration of the current round.

[0080] After step S23, the following is also included:

[0081] If no data is received from the target neighbor node after the preset waiting period, it is determined that a communication packet loss has occurred, and the message variance corresponding to the target neighbor node is forcibly assigned to a preset maximum constant, and its message mean is forcibly assigned to zero.

[0082] As an example, to address the instability of underlying communication networks in harsh environments, nodes... Execute packet loss detection and replacement logic. Node A preset communication waiting time window is opened (in this configuration example, the window length is set to 50 milliseconds; mainstream power distribution terminals (such as fiber optic or 5G / 4G wireless communication) only require a few milliseconds to a dozen milliseconds for inbound transmission in a healthy state, and 50ms is sufficient to cover normal communication jitter). If the node Successfully received neighboring nodes within the communication waiting time window If the data packet is invalid, then extract the data normally. and Participate in subsequent calculations.

[0083] If node If no data packet is received from the target neighbor node after the communication waiting time window expires, the system determines that packet loss has occurred. At this time, the node... Automatically perform default value replacement operation: replace the missing target neighbor node (with... (For example) message variance The forced assignment is a preset maximum constant that is adaptive to the local units (e.g., the value is taken as the local observation variance of the node). 10 4 (times), and average the corresponding messages. Forced to be assigned a value of 0.

[0084] Step S24 combines the initial local observation mean, local observation variance, message mean, and message variance to obtain the expected load assessment value and data missingness, specifically including:

[0085] The local observation variance of the current node is combined with the message variance of all neighboring nodes in the previous round to obtain the comprehensive accuracy parameter of the current iteration round.

[0086] As an example, the current node After receiving data from neighboring nodes (or completing packet loss replacement processing), the core numerical simulation phase begins. First, the nodes... Based on the received messages from neighboring nodes, calculate the first... Overall accuracy parameters of the wheel The calculation method is as follows:

[0087]

[0088] in: Represents a node The local observation variance; Indicates adjacent nodes in the previous round Send to node The message variance; Represents the set of adjacent communication nodes Sum all adjacent nodes; single item and It represents the reciprocal of the corresponding variance, and in statistics, it represents the precision of the data.

[0089] This formula indicates that the improvement in node integration accuracy depends on the influx of valid data from the surrounding area. When packet loss occurs, the system will... Assign a value that is a very large constant (e.g., This causes its reciprocal term to approach 0, at which point the contribution of the missing node to the summation term disappears, resulting in a decrease in the calculated overall accuracy parameter. The significant reduction objectively reflects the decrease in accuracy caused by missing data.

[0090] Based on the comprehensive accuracy parameter, the initial local observation mean, the local observation variance, and the weighted sum of the message mean and message variance of all adjacent nodes, the expected load assessment value for the current iteration round is calculated.

[0091] As an example, the expected load assessment value represents the projected average of the expected power data for the k-th round, and the current node... Expected load assessment value The calculation method is as follows:

[0092]

[0093] in: Represents a node The initial local observation mean; Indicates the local observation variance. Indicates adjacent nodes in the previous round Send to node The average message value; This represents a weighted summation of local observation data and input data from all neighboring nodes, based on their respective precision. The formula indicates that the projected mean is a smoothed summary of the network's overall effective load information; the smaller the variance of a node's message (the more reliable the data), the higher its contribution to the final mean. The greater the traction effect, the more distributed data can be integrated without relying on a central controller.

[0094] The data missingness of the current iteration is determined based on the reciprocal of the comprehensive accuracy parameter.

[0095] As an example, the current node The Data missing degree of round The calculation formula is as follows:

[0096]

[0097] in: This represents the overall accuracy parameter calculated above; Represents a node The data missingness (i.e., communication missing variance) assessed in this round. The formula indicates the data missingness... With comprehensive accuracy parameters It exhibits a strict inverse proportional relationship. Combining this with the aforementioned accuracy pattern, it can be seen that the more severe the packet loss at surrounding nodes, the higher the accuracy. The smaller the value, the lower the data missingness in the final output. It will grow drastically.

[0098] The loop terminates when the number of iterations reaches the preset maximum number of iterations, and the final expected load assessment value and data missingness are output.

[0099] As an example, in order for the deduction to propagate continuously within the network, nodes must transmit their updated states outwards, forming an iterative closed loop. After calculating the... Expected load assessment value of the wheel With data missing degree After that, node This is directly mapped to message variables that are passed outwards. Node Set the next round to send to all adjacent nodes message average Set message variance .

[0100] Subsequently, the node Send the updated message mean and message variance to the set of neighboring communication nodes. This is to support the neighboring nodes in the next round of deduction. Furthermore, the nodes... Check the current iteration round Has the maximum number of iterations been reached? If it is not achieved, then... And continue the iterative process; if the maximum number of iterations is reached... Then the node Terminate the loop. Output the final expected load assessment and data missing rate.

[0101] Step S30: Based on the expected load assessment value and the data missingness, calculate the risk level of cable overload at the communication node, specifically including:

[0102] The expected load assessment value and the data missingness are weighted and summed to obtain the risk level of cable overload at the communication node.

[0103] As an example, the risk level value of node i The calculation formula is:

[0104]

[0105] in: This represents the standard deviation of the missing data, used to reduce the dimensionality of the missing data to the expected workload assessment value. Numerical values ​​with the same dimensions; This represents the preset dimensionless engineering transformation weight (in this configuration example, it can be uniformly set to 1.0), which is used to smoothly map data uncertainty into a capacity penalty term; Indicates the coefficient of sudden increase in complex current;

[0106] This represents the capacity penalty compensation item, which translates the risk of packet loss into a specific reserve load value; This indicates the maximum expected capacity after considering the worst-case communication failure and the impact of equipment startup. Indicates the rated thermal limit capacity of the cable;

[0107] This represents the calculated risk level of cable overload at the node. The formula indicates the degree of data missing. The larger the value (indicating more severe packet loss in the surrounding area leading to more missing data), and the greater the power-on surge coefficient. The larger the value (indicating more inductive devices such as local motors and a more severe power-on surge), the larger the calculated capacity penalty compensation value will be. Adding this value to the mean and dividing by the denominator of the physical thermal limit yields the final risk level value. This will increase proportionally. This value objectively and conservatively quantifies the actual risk of the cable exceeding its rated load limit at the moment of power restoration at the current node, fundamentally preventing overload due to communication interruptions. The cable overload assessment value for this node is then output for hierarchical comparison across the entire network.

[0108] Before step S40, the following steps are also included:

[0109] When the maximum value of the risk level does not exceed the preset safety threshold, the control switch will close.

[0110] As an example, after completing the extreme value assessment of any node, the discrete risk data across the entire network is reduced to a single extreme value required by the control center. To avoid network congestion caused by flooding, nodes adopt a unicast backhaul method based on upstream addresses, specifically:

[0111] Taking node i as an example, if the probe frame first received by a neighboring node is forwarded from node i, that neighboring node will register node i as its upstream node, becoming a child node of node i. The child node is responsible for unicasting its calculated local overload assessment extreme value to node i. Node i, acting as the aggregation hub, is responsible for collecting data from all child nodes, comparing it with the local overload extreme value, selecting the maximum value, and then forwarding it to its own upstream node. After a neighboring child node completes its local assessment calculation, the child node uses the upstream communication node address it registered in the above steps. The calculated risk level value is encapsulated along with its own node network address into a child node evaluation value data packet. The child node then performs a unicast transmission to that upstream address.

[0112] node Currently in receiving mode. Node By receiving the aforementioned unicast messages, the risk level values ​​returned by all adjacent child nodes considered upstream are obtained. The risk level value calculated locally The numerical value is compared with the evaluation values ​​of all received child nodes. Node The maximum value among these values ​​is selected and taken as the convergent extremum for this region. Meanwhile, the nodes... Extract the Media Access Control (MAC) address of the node that generated this maximum value. The convergence extreme value and its corresponding MAC address are repackaged to generate a line overload assessment data packet.

[0113] Ultimately, the node Call the upstream communication node address registered in the above steps. The overload assessment data packet for this line is then forwarded to the next higher-level node.

[0114] As an example, the preset safety threshold is the system safety boundary constant. (For example, this constant is fixed at 1.0, representing the physical upper limit of the trunk cable's rated heat dissipation). The tie switch receives the final converged line overload assessment data packet. The tie switch parses this data packet and extracts the maximum value of the risk level value. And the address of the evaluation node that caused this extreme value. .

[0115] Furthermore, the interconnecting switch executes numerical comparison logic to determine the maximum value of the risk level. Is it greater than the system safety boundary constant? Where the extreme value is much smaller than the system safety boundary constant. This indicates that even with the combined effects of a surge in demand during equipment cold starts and worst-case communication packet loss, the increased capacity of the most vulnerable nodes in the current power outage network is still within the safety margin of the backbone cables. Conversely, if this extreme value exceeds the system safety boundary constant... This indicates that the transient surge current of the local network after power restoration must exceed the carrying capacity limit of the main cable, and direct closing of the circuit breaker will trigger a secondary trip.

[0116] As an example, when the risk level value is at its maximum value Less than or equal to the system safety boundary constant When the tie switch determines that the remaining network topology is fully capable of withstanding the power restoration impact, the controller inside the tie switch directly issues a closing level command to the local switch operating mechanism. The mechanical contacts of the tie switch complete the closing action, and the power from the backup line flows into the power outage area, allowing the entire network to smoothly and safely restore power.

[0117] Step S40: When the maximum value of the risk level exceeds the preset safety threshold, the target node is controlled to disconnect the local load access side circuit breaker, and the tie switch is controlled to close after isolating the over-limit load. The target node is the node corresponding to the maximum value of the risk level.

[0118] As an example, when the risk level value is at its maximum value Greater than the system safety boundary constant At that time, the tie switch determined that there was a risk of exceeding the limit when closing the entire circuit. In order to ensure the power supply to most of the healthy network, the tie switch controller called the well-known fixed-point message communication protocol of the distribution network, using the address of the node that was evaluated for exceeding the limit extracted from the data packet. For the target address, a branch disconnect command is issued to that node. The target node prioritizes controlling the circuit breakers on the local load access side under its jurisdiction to perform disconnection actions, thereby cutting off the high-heat, heavy-load equipment that caused the overload to exceed the limit.

[0119] Step S40, which involves controlling the target node to disconnect the local load access side circuit breaker, includes:

[0120] A branch disconnection command is sent to the target node via the tie switch. After receiving the branch disconnection command, the target node controls the local load access side circuit breaker to disconnect and execute the disconnection operation.

[0121] As an example, the tie switch uses the extracted target node network address as the destination address, encapsulates and sends a directed branch disconnect message. The target node matches the address and receives the message, parses it to obtain the branch disconnect command, and the target node controller outputs a trip trigger level to the operating mechanism of the local load access side circuit breaker. The local load access side circuit breaker performs a mechanical disconnect action, physically isolating the over-limit local load.

[0122] After step S40, the following steps are also included:

[0123] The control target node maintains the circuit breakers on both sides of the main line in a closed and connected state to ensure the smooth transmission of main line power to downstream healthy nodes.

[0124] As an example, the target node forcibly maintains the circuit breakers on both sides of its main line in the closed state to ensure the connectivity of the main distribution network topology. The target node sends a confirmation frame of completion back to the tie switch, triggering the tie switch to perform the subsequent main line closing operation. After confirming that the overload of the target node has been successfully isolated, the tie switch controller sends a closing level command to the local switch operating mechanism, and the tie switch performs the closing action. Since the most dangerous overload in the system has been accurately eliminated, the closing of the tie switch allows the remaining healthy network to safely restore power supply. If multiple nodes have overload risks, after the current operation, a multi-agent interactive evaluation is retried until the maximum value of the risk level after the re-evaluation does not exceed the preset safety threshold before the tie switch is controlled to close.

[0125] This application provides a self-healing control method for distribution network faults based on multi-agent cooperation. It iteratively interacts and outputs data missingness using a Gaussian belief propagation algorithm. This feature maintains node inference calculations even when data packets are lost in the communication network, avoiding computational stagnation caused by waiting for missing data in traditional distributed summation algorithms. The method combines the expected load assessment value with the data missingness to calculate the cable overload risk level. Then, based on the maximum risk level exceeding the limit, it controls the target node circuit breaker to open and the tie switch to close. This performs targeted load shedding based on the risk level value, reducing the control center's dependence on complete global data and avoiding situations where blind closing can easily lead to mainline cable overload and cascading tripping.

[0126] like Figure 2 In another embodiment of the present invention, an electronic device is also provided, comprising:

[0127] Memory 101 is used to store one or more computer programs;

[0128] Processor 102 is configured to execute a computer program to implement the steps of a multi-agent cooperative distribution network fault self-healing control method as described in any of the above embodiments.

[0129] In this embodiment, the computer system suitable for implementing the electronic device of the present invention includes a Central Processing Unit (CPU), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) or a program loaded from storage into random access memory (RAM), such as performing the methods described in the above embodiments. Various programs and data required for system operation are also stored in RAM. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0130] The following components are connected to the I / O interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard drives; and communication sections including network interface cards such as LAN (Local Area Network) cards and modems. The communication sections perform communication processing via networks such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.

[0131] The specific implementation method of the distribution network fault self-healing control device based on multi-agent cooperation in this application is basically the same as the embodiments of the distribution network fault self-healing control method based on multi-agent cooperation described above, and will not be repeated here.

[0132] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system 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 system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0133] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0134] The above are merely preferred embodiments of this application and do not limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

Claims

1. A self-healing control method for distribution network faults based on multi-agent cooperation, characterized in that, The method includes: The steady-state power data of multiple communication nodes in the distribution network before the fault trip is obtained. Based on the steady-state power data and the power outage duration, the expected power data of each communication node at the moment of power restoration is calculated. The communication nodes interact with each other through an intelligent agent. Based on the expected power data and the Gaussian belief propagation algorithm, message iteration processing is performed on adjacent communication nodes to obtain the expected load assessment value and data missingness of the communication nodes. Based on the expected load assessment value and the data missingness, the risk level of cable overload at the communication node is calculated. When the maximum value of the risk level exceeds the preset safety threshold, the target node is controlled to disconnect the local load access side circuit breaker, and the tie switch is controlled to close after isolating the over-limit load. The target node is the node corresponding to the maximum value of the risk level.

2. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, The calculation of the expected power data of each communication node at the moment of power restoration, based on the steady-state power data and the power outage duration, includes: Determine the power restoration surge coefficient and the power outage cooling coefficient; Based on the power restoration surge coefficient, power outage cooling coefficient, and power outage duration, the surge multiple of the power data relative to the steady-state power data at the moment of power restoration is calculated. Based on the surge factor and the steady-state power data, the expected power data of each communication node at the moment of power restoration is calculated.

3. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, The process of iteratively processing messages between adjacent communication nodes based on the expected power data and the Gaussian belief propagation algorithm to obtain the expected load assessment value and data missingness of the communication nodes includes: For any communication node, the expected power data is used as the initial local observation mean of the current node in the Gaussian belief propagation algorithm; The local observation variance is calculated based on the square of the initial local observation mean. In any iteration of the Gaussian belief propagation algorithm, the mean and variance of messages sent by each neighboring node are received; The initial local observation mean, local observation variance, message mean, and message variance are combined to obtain the expected load assessment value and data missingness.

4. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 3, characterized in that, The step of combining the initial local observation mean, local observation variance, message mean, and message variance to obtain the expected load assessment value and data missingness includes: The local observation variance of the current node is combined with the message variance of all neighboring nodes in the previous round to obtain the comprehensive accuracy parameter of the current iteration round. Based on the comprehensive accuracy parameter, the initial local observation mean, the local observation variance, and the weighted sum of the message mean and message variance of all adjacent nodes, the expected load assessment value for the current iteration round is calculated. The data missingness of the current iteration is determined based on the reciprocal of the comprehensive accuracy parameter. The loop terminates when the number of iterations reaches the preset maximum number of iterations, and the final expected load assessment value and data missingness are output.

5. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 3, characterized in that, After receiving the mean and variance of messages sent by each adjacent node, the method further includes: If no data is received from the target neighbor node after the preset waiting period, it is determined that a communication packet loss has occurred, and the message variance corresponding to the target neighbor node is forcibly assigned to a preset maximum constant, and its message mean is forcibly assigned to zero.

6. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, The calculation of the risk level of cable overload at the communication node based on the expected load assessment value and the data missingness includes: The expected load assessment value and the data missingness are weighted and summed to obtain the risk level of cable overload at the communication node.

7. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, Before controlling the target node to disconnect the local load access side circuit breaker when the maximum value of the risk level exceeds the preset safety threshold, the procedure further includes: When the maximum value of the risk level does not exceed the preset safety threshold, the control switch is closed.

8. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, The control target node disconnects the local load access side circuit breaker, including: A branch disconnection command is sent to the target node via the tie switch. After receiving the branch disconnection command, the target node controls the local load access side circuit breaker to disconnect and execute the disconnection operation.

9. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, After the control target node disconnects the local load access side circuit breaker, it also includes: The control target node maintains the circuit breakers on both sides of the main line in a closed and connected state to ensure the smooth transmission of main line power to downstream healthy nodes.

10. The self-healing control method for distribution network faults based on multi-agent cooperation as described in claim 1, characterized in that, The acquisition of steady-state power data of multiple communication nodes in the distribution network before fault tripping includes: Acquire active power and reactive power data of multiple communication nodes in the distribution network within a preset steady-state time period before fault tripping; The active power data and reactive power data are integrated and processed to obtain steady-state power data.

Images