A method for locating safe faults in active distribution networks based on integrated synaesthesia and computation

Abstract

The present invention relates to the field of active distribution network safety, and discloses a method for locating active distribution network safety faults based on integrated synaesthesia and arithmetic. The method uses sensors to sense the direction of fault current in the active distribution network and then transmits the monitoring and control system to convert the fault current information into the section state information of the active distribution network, and judges the accuracy of the fault information by calculating the error function, and finally realizes the ability to quickly and accurately locate faults according to environmental changes through the training of a meta-reinforcement learning model.

Description

Technical Field

[0001] The invention relates to the field of active power distribution network safety, and discloses an active power distribution network safety fault positioning method based on integrated synaesthesia and calculation. Background Art

[0002] More and more distributed power sources are connected to the power system, which has changed the power supply mode of the traditional distribution network, transforming it from a single power supply to a multi-power supply. This change makes the topology of the distribution network and the equipment operating environment more complicated. The distribution network is the intermediate link between the power grid and the user. During the transmission process, it is susceptible to the influence of bad weather and irresistible factors, which may cause the distribution network to fail. If the fault is not handled in time, it may affect the stable operation of the power system and have an adverse impact on people's living standards. Therefore, how to quickly and accurately determine the location of the fault point has become an urgent problem to be solved in the fault location of the distribution network.

[0003] The traditional fault location technologies for the access of distributed power sources, such as the implementation of remote monitoring systems, infrared thermal imaging technology, and intelligent sensor networks, require a large amount of equipment investment and technical support, resulting in high costs and failing to meet today's needs. The introduction of the integrated sensing and computing technology achieves a trade-off between perception, communication, and computing on the same platform, saving resources and reducing costs. It combines perception and communication with computing power for real-time analysis to achieve rapid feedback on the system status, and analyzes possible errors in the perception, communication, and computing process, thereby improving the overall safety performance of the system.

[0004] Intelligent algorithms are indispensable in fault location of active distribution networks. Classical whale algorithms, gray wolf algorithms, ant colony algorithms, etc. cannot make decisions quickly and adaptively according to changes in the environment, and have slow convergence speed and are greatly affected by initial values. Meta-reinforcement learning algorithms can solve the above problems due to their strong generalization ability, fast learning speed, and adaptability to diversified tasks. Therefore, this paper finally optimizes system performance through meta-reinforcement learning algorithms. Summary of the invention

[0005] The present invention aims to solve one of the technical problems existing in the prior art or related technology. A method for locating safe faults in an active distribution network based on integrated telepathy and computing is proposed. The method can calculate corresponding data indicators while sensing and communicating through the mutual integration of telepathy and computing technologies, save resources, and improve efficiency. By integrating meta-reinforcement learning, the fault location can still be quickly, accurately and safely located when the active distribution network environment changes.

[0006] The embodiment of the present invention provides the following technical solutions:

[0007] The present invention provides a method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation, the method comprising:

[0008] A method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation, characterized in that it comprises the following steps:

[0009] Step A: Sensing and describing fault current at active distribution network nodes.

[0010] Step B, calculate the probability of reliable transmission of data information, and convert it into the status information of the entire distribution network section based on the current information of the fault node.

[0011] Step C: constructing an error function of the fault information accuracy determined by the conversion function.

[0012] Step D: Based on the calculation results of steps A, B, and C, a meta-reinforcement learning algorithm model is constructed to improve the accuracy of distribution network fault location.

[0013] Wherein, step A specifically includes:

[0014] A1, the sensing terminal in the active distribution network is equipped with N intelligent sensors to sense the fault current information of N feeder nodes respectively.

[0015] A2, the sensed fault current flow direction information is expressed as:

[0016] The positive direction of the fault current is defined as the direction from the fixed power source to the end of the feeder or the distributed power source. Indicates that the fault current flows in the same direction as the positive direction. Indicates that no fault current flows. Indicates that the fault current flows in the opposite direction to the positive direction. , This indicates that the node has failed.

[0017] A3, define the nth ( The signal received by each sensor for judging the direction of the fault current is:

[0018]

[0019] in, is the sensing signal sent by the sensor. is the reflection coefficient, is the angle of arrival, and are the transmit and receive steering vectors, respectively.

[0020] A4, calculate the false detection probability of the nth sensor:

[0021]

[0022] in , are the threshold ranges of arrival angle and reflection coefficient respectively.

[0023] Wherein, step B specifically includes:

[0024] B1, transmits the sensed fault current information to the monitoring and control system, and calculates the reliable transmission probability of communication:

[0025]

[0026] in, To find the mean, , is the communication signal-to-noise ratio, is the communication information transmission rate, V is the channel divergence, and M is the block length of the transmitted information.

[0027] B2, the state information of the feeder section of the active distribution network is determined according to the fault current information at the node. Therefore, the node fault current information needs to be converted into feeder section information through a switching function. The constructed switching function is:

[0028]

[0029]

[0030]

[0031] in For the OR operation, is the mean value of the state information of node f, is the mean state information of the upstream part of node f, is the mean of the downstream partial state information of node f. are the number of upstream power sources, the number of downstream power sources, the number of upstream sections of node f, and the number of downstream sections of node f, respectively. are the switching coefficients of the upstream and downstream power supplies of node f, respectively. The value is 1 when the power supply is connected and 0 when the power supply is disconnected. is the OR value of the mean state of the section between node f and the upstream power supply, is the mean or value of the state of all sections downstream of node f, is the OR value of the mean state of the segment between node f and the downstream power supply, It is the sum of the state means of all segments upstream of node f.

[0032] Wherein, step C specifically includes:

[0033] C1, the error function is used to describe the difference between the distribution network fault status information calculated by the switching function in step B2 and the distribution network fault status monitoring information obtained by the sensors on each node. The smaller the difference, the closer the information calculated is to the actual situation.

[0034] C2, the constructed error function is:

[0035]

[0036] Where N is the total number of distribution network nodes, K is the total number of feeder sections, For the OR operation, is the proportionality coefficient.

[0037] Wherein, step D specifically comprises:

[0038] D1, each fault location of the active distribution network is modeled as a Markov decision process (MDP), which contains the state space , Action Space , reward function , transition probability and the end signal Composition, expressed in the form of a quintuple: .

[0039] The state space , They represent the fault conditions of the nth node and the kth section of the active distribution network respectively; the action space , are the false detection probability of the nth sensor and the reliable transmission probability of communication; the reward function ; , Indicates the end of the process. Indicates that the process continues.

[0040] D2, each MDP process constitutes a meta-task of meta-reinforcement learning, so the meta-task set is expressed as: , represents the i-th meta-task.

[0041] D3, the meta-reinforcement learning optimization process is divided into two stages: meta-training and meta-testing. In the meta-training stage, the task ( The training set (including training set and test set) is sent to the training model for training, and the loss is obtained. , use this loss to update the initial parameters of the model , and the updated model parameters are: ,in is the inner loop learning rate, The parameter is The fault location model is based on the parameters Based on the test set, the model is tested to obtain the task The loss is:

[0042]

[0043] Therefore, after the inner loop of i tasks, the loss function is accumulated to form the loss of all tasks:

[0044]

[0045] Secondly, the initial parameters of the model are updated through the outer loop of meta-training have to for:

[0046]

[0047] in, is the learning rate of the outer loop, and the objective function of the meta-training phase is .

[0048] D4, meta-testing phase Based on the results of meta-training, a new task is sampled in the meta-task set for testing to obtain the optimized parameters of the final model: .

[0049] D5, through continuous iterative training of the model, the accuracy, safety and reliability of fault location in the active distribution network are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 The present invention is a flowchart of a method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation. DETAILED DESCRIPTION

[0052] As described in the background technology section, how to ensure the safety and accuracy of fault location in active distribution networks, realize the integrated functions of system communication network, perception network and computing functions, and how to achieve adaptive and accurate fault location in a complex distribution network environment with distributed power sources are urgent problems to be solved by people in this field.

[0053] The core idea of ​​the present invention is that, through the integrated technology of tele-sensing and computing, when a fault occurs in the active distribution network, the sensing terminal timely collects the fault current information and transmits it wirelessly to the monitoring and control platform for further analysis and calculation, which is converted into the feeder section status information of the distribution network, and the positioning accuracy is improved through the meta-reinforcement learning algorithm.

[0054] Reference Figure 1 The present invention provides a method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation, the method comprising:

[0055] A method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation, characterized in that it comprises the following steps:

[0056] Step A: Sensing and describing fault current at active distribution network nodes.

[0057] Step B, calculate the probability of reliable transmission of data information, and convert it into the status information of the entire distribution network section based on the current information of the fault node.

[0058] Step C: constructing an error function of the fault information accuracy determined by the conversion function.

[0059] Step D: Based on the calculation results of steps A, B, and C, a meta-reinforcement learning algorithm model is constructed to improve the accuracy of distribution network fault location.

[0060] Wherein, step A specifically includes:

[0061] A1, the sensing terminal in the active distribution network is equipped with N intelligent sensors to sense the fault current information of N feeder nodes respectively.

[0062] A2, the sensed fault current flow direction information is expressed as:

[0063] The positive direction of the fault current is defined as the direction from the fixed power source to the end of the feeder or the distributed power source. Indicates that the fault current flows in the same direction as the positive direction. Indicates that no fault current flows. Indicates that the fault current flows in the opposite direction to the positive direction. , This indicates that the node has failed.

[0064] A3, define the nth ( The signal received by each sensor for judging the direction of the fault current is:

[0065]

[0066] in, is the sensing signal sent by the sensor. is the reflection coefficient, is the angle of arrival, and are the transmit and receive steering vectors, respectively.

[0067] A4, calculate the false detection probability of the nth sensor:

[0068]

[0069] in , are the threshold ranges of arrival angle and reflection coefficient respectively.

[0070] Wherein, step B specifically includes:

[0071] B1, transmits the sensed fault current information to the monitoring and control system, and calculates the reliable transmission probability of communication:

[0072]

[0073] in, To find the mean, , is the communication signal-to-noise ratio, is the communication information transmission rate, V is the channel divergence, and M is the block length of the transmitted information.

[0074] B2, the state information of the feeder section of the active distribution network is determined according to the fault current information at the node. Therefore, the node fault current information needs to be converted into feeder section information through a switching function. The constructed switching function is:

[0075]

[0076]

[0077]

[0078] in For the OR operation, is the mean value of the state information of node f, is the mean state information of the upstream part of node f, is the mean of the downstream partial state information of node f. are the number of upstream power sources, the number of downstream power sources, the number of upstream sections of node f, and the number of downstream sections of node f, respectively. are the switching coefficients of the upstream and downstream power supplies of node f, respectively. The value is 1 when the power supply is connected and 0 when the power supply is disconnected. is the OR value of the mean state of the section between node f and the upstream power supply, is the mean or value of the state of all sections downstream of node f, is the OR value of the mean state of the segment between node f and the downstream power supply, It is the sum of the state means of all segments upstream of node f.

[0079] Wherein, step C specifically includes:

[0080] C1, the error function is used to describe the difference between the distribution network fault status information calculated by the switching function in step B2 and the distribution network fault status monitoring information obtained by the sensors on each node. The smaller the difference, the closer the information calculated is to the actual situation.

[0081] C2, the constructed error function is:

[0082]

[0083] Where N is the total number of distribution network nodes, K is the total number of feeder sections, For the OR operation, is the proportionality coefficient.

[0084] Wherein, step D specifically comprises:

[0085] D1, each fault location of the active distribution network is modeled as a Markov decision process (MDP), which contains the state space , Action Space , reward function , transition probability and the end signal Composition, expressed in the form of a quintuple: .

[0086] The state space , They represent the fault conditions of the nth node and the kth section of the active distribution network respectively; the action space , are the false detection probability of the nth sensor and the reliable transmission probability of communication; the reward function ; , Indicates the end of the process. Indicates that the process continues.

[0087] D2, each MDP process constitutes a meta-task of meta-reinforcement learning, so the meta-task set is expressed as: , represents the i-th meta-task.

[0088] D3, the meta-reinforcement learning optimization process is divided into two stages: meta-training and meta-testing. In the meta-training stage, the task ( The training set (including training set and test set) is sent to the training model for training, and the loss is obtained. , use this loss to update the initial parameters of the model , and the updated model parameters are: ,in is the inner loop learning rate, The parameter is The fault location model is based on the parameters Based on the test set, the model is tested to obtain the task The loss is:

[0089] Therefore, after the inner loop of i tasks, the loss function is accumulated to form the loss of all tasks:

[0090]

[0091] Secondly, the initial parameters of the model are updated through the outer loop of meta-training have to for:

[0092]

[0093] in, is the learning rate of the outer loop, and the objective function of the meta-training phase is .

[0094] D4, meta-testing phase Based on the results of meta-training, a new task is sampled in the meta-task set for testing to obtain the optimized parameters of the final model: .

[0095] D5, through continuous iterative training of the model, the accuracy, safety and reliability of fault location in the active distribution network are improved.

[0096] Compared with the prior art, the above technical solution has the following advantages:

[0097] The present invention is based on a safe fault location method for an active distribution network based on integration of communication, sensing and computing. The technology of integration of communication, sensing and computing is introduced into the fault location method for the active distribution network. The sensing terminal senses the fault current information of the distribution network and calculates the sensed error value at the same time. In the process of transmitting the fault current information to the monitoring and control system, the safe communication rate is calculated and the sensed data is converted into the distribution network feeder fault section information. At the same time, a meta-reinforcement learning algorithm is introduced to optimize the error in the sensed communication process, so that the active distribution network can quickly make safe and accurate fault location according to environmental changes. Compared with the technology of fault location using traditional methods, the present invention not only improves the safety and accuracy of the system but also reduces the cost. Therefore, the present invention can better meet the current needs.

[0098] The present invention discloses a method for locating safe faults in an active distribution network based on integrated synaesthesia and calculation. The method uses sensors to sense the direction of fault current in the active distribution network and then transmits the monitoring and control system to convert the fault current information into the section status information of the active distribution network, and judges the accuracy of the fault information by calculating the error function. Finally, the method can quickly and accurately locate the fault according to environmental changes through the training of the meta-reinforcement learning model.

[0099] The various parts in this manual are described in a progressive manner, and each part focuses on the differences from other parts. The same or similar parts between the various parts can be referenced to each other.

[0100] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for locating safe faults in active distribution networks based on integrated synaesthesia and calculation, characterized in that: The steps include: Step A, sensing and description of fault current at active distribution network nodes; Step B, calculate the probability of reliable transmission of data information, and convert it into the status information of the entire distribution network section according to the current information of the fault node; Step C, constructing an error function of the fault information accuracy determined by the conversion function; Step D, based on the calculation results of steps A, B, and C, a meta-reinforcement learning algorithm model is constructed to improve the accuracy of distribution network fault location; Wherein, step A specifically includes: A1, the sensing terminal in the active distribution network is equipped with N intelligent sensors to sense the fault current information of N feeder nodes respectively; A2, the sensed fault current flow direction information is expressed as: ; The positive direction of the fault current is defined as the direction from the fixed power source to the end of the feeder or the distributed power source. Indicates that the fault current flows in the same direction as the positive direction. Indicates that no fault current flows. Indicates that the fault current flows in the opposite direction to the positive direction. , All indicate that this node has failed; A3, define the nth ( The signal received by each sensor for judging the direction of the fault current is: ; in, is the perception signal sent by the sensor, is the reflection coefficient, is the angle of arrival, and are the transmitting and receiving steering vectors respectively; A4, calculate the false detection probability of the nth sensor: ; in , are the threshold ranges of arrival angle and reflection coefficient respectively; Wherein, step B specifically includes: B1, transmits the sensed fault current information to the monitoring and control system, and calculates the reliable transmission probability of communication: ; in, To find the mean, , is the communication signal-to-noise ratio, is the communication information transmission rate, V is the channel divergence, and M is the block length of the transmitted information; B2, the state information of the feeder section of the active distribution network is determined according to the fault current information at the node. Therefore, the node fault current information needs to be converted into feeder section information through a switching function. The constructed switching function is: , , ; in For the OR operation, is the mean value of the state information of node f, is the mean state information of the upstream part of node f, is the mean of the downstream partial state information of node f, are the number of upstream power sources, the number of downstream power sources, the number of upstream sections of node f, and the number of downstream sections of node f, respectively. are the switching coefficients of the upstream and downstream power supplies of node f, respectively. The value is 1 when the power supply is connected and 0 when the power supply is disconnected. is the OR value of the mean state of the section between node f and the upstream power supply, is the mean or value of the state of all sections downstream of node f, is the OR value of the mean state of the segment between node f and the downstream power supply, is the OR value of the mean state of all segments upstream of node f; Wherein, step C specifically includes: C1, the error function is used to describe the difference between the distribution network fault status information calculated by the switching function in step B2 and the distribution network fault status monitoring information obtained by the sensor monitoring on each node. The smaller the difference, the closer the information calculated is to the actual situation. C2, the constructed error function is; ; Where N is the total number of distribution network nodes, K is the total number of feeder sections, For the OR operation, is the proportionality coefficient; Wherein, step D specifically comprises: D1, each fault location of the active distribution network is modeled as a Markov decision process (MDP), which contains the state space , Action Space , reward function , transition probability and the end signal Composition, expressed in the form of a quintuple: ; The state space , They represent the fault conditions of the nth node and the kth section of the active distribution network respectively; the action space , are the false detection probability of the nth sensor and the reliable transmission probability of communication; the reward function ; , Indicates the end of the process. Indicates that the process continues; D2, each MDP process constitutes a meta-task of meta-reinforcement learning, so the meta-task set is expressed as: , represents the i-th meta-task; D3, the meta-reinforcement learning optimization process is divided into two stages: meta-training and meta-testing. In the meta-training stage, the task ( The training set (including training set and test set) is sent to the training model for training, and the loss is obtained. , use this loss to update the initial parameters of the model , and the updated model parameters are: ,in is the inner loop learning rate, The parameter is The fault location model is based on the parameters Based on the test set, the model is tested to obtain the task The loss is: ; Therefore, after the inner loop of i tasks, the loss function is accumulated to form the loss of all tasks: ; Secondly, the initial parameters of the model are updated through the outer loop of meta-training have to for: ; in, is the learning rate of the outer loop, and the objective function of the meta-training phase is ; D4, meta-testing phase Based on the results of meta-training, a new task is sampled in the meta-task set for testing to obtain the optimized parameters of the final model: ; D5, through continuous iterative training of the model, the accuracy, safety and reliability of fault location in the active distribution network are improved.

Images