Smart grid fault location and self-healing control optimization system

Abstract

The application relates to the technical field of power system automation, and discloses a smart power grid fault positioning and self-healing control optimization system, which comprises a multi-source time sequence evidence gathering module, a topology-parameter online self-calibration module, a fault hypothesis set generation module, a belief updating module, an active detection action planning module, a risk-restricted self-healing joint optimization module, an online safety checking and rollback module and a closed-loop execution module. Through the adoption of the control action of the controllable equipment of the power grid into the decision link of fault positioning, the power grid response data generated after the control instruction is issued is returned to the multi-source time sequence evidence gathering module and the observation vector is reconstructed, the module is driven again, the online safety checking and rollback module executes a new round of solving, a closed loop of positioning-control-observation-repositioning is formed, logical disconnection between the positioning result and the self-healing control is avoided, and it is ensured that each control action participates in the evidence updating process of the confidence convergence of the fault position.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, specifically to a smart grid fault location and self-healing control optimization system. Background Technology

[0002] Existing smart grid fault handling systems typically employ a serial technical framework, which first infers the fault location through sensor measurements or protection action data, and then performs fault isolation and network reconfiguration recovery control based on the determined fault point. Such systems rely on the passive acquisition of continuous measurement data and event-based data, and output a single fault conclusion through an independent fault location algorithm, which is then used as a pre-input for a self-healing control system.

[0003] In the aforementioned serialized framework, control actions and fault location lack a unified terminology system and confidence evolution link. The power grid response data generated after the system executes isolation or network reconfiguration control commands does not participate in the confidence update closed loop of fault location, resulting in a logical disconnect between control action issuance and fault location. This prevents the control results from flowing back to the fault location data link, thus making it impossible to establish a closed loop of location-control-observation-relocation. The integrated closed-loop decision-making logic of control actions and fault location is missing, making it difficult to support system-level fault location confidence convergence. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a smart grid fault location and self-healing control optimization system. This system solves the problem that in existing systems, the grid response data generated after executing isolation or network reconfiguration control commands does not participate in the confidence update closed loop of fault location, resulting in a logical disconnect between control action issuance and fault location, and preventing control results from flowing back to the fault location data link.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a smart grid fault location and self-healing control optimization system, comprising: The module includes a multi-source temporal evidence aggregation module, a topology-parameter online self-calibration module, a fault hypothesis set generation module, a belief update module, an active detection action planning module, a risk-constrained self-healing joint optimization module, an online safety verification and rollback module, and a closed-loop execution module. The multi-source time-series evidence aggregation module outputs an observation vector. The online self-calibration module for topology parameters outputs a feasible set of network parameters. The fault hypothesis set generation module outputs a fault hypothesis set; The belief update module forms beliefs based on the observation vector and the fault hypothesis set under the constraints of the feasible set of network parameters; The active detection action planning module generates a set of candidate control actions based on the belief and calculates the information gain. The risk-constrained self-healing joint optimization module uses the power restoration cost, action cost, and belief entropy of the belief to form a joint objective, and solves the control action vector of the rolling plan from the candidate control action set under the constraints of power grid operation and risk. The online security verification and rollback module performs a feasibility verification of the control action vector based on the set of credible fault assumptions determined by the belief and the set of feasible network parameters, and generates a rollback control action when the verification fails. The closed-loop execution module executes the verified control actions and writes back continuous measurement data and event data to update the observation vector to form a closed loop.

[0006] Preferably, the control action vector includes control actions at multiple times, and the control actions at each time are represented by a consistent structure and serve as constituent elements of the control action vector; wherein, each control action simultaneously includes: changes in the reactive power setting of the distributed power source, changes in the switching of reactive power compensation, changes in the tap changer of the voltage regulating transformer, and changes in the switch state; and, each candidate control action in the candidate control action set generated by the active detection action planning module satisfies the corresponding action boundary constraints.

[0007] Preferably, the active detection action planning module determines the information gain according to the following rules: The difference between the belief entropy before the execution of the candidate control action and the expected belief entropy under the predicted observation distribution after the execution of the candidate control action is used as the information gain of the candidate control action. The candidate control action set is then filtered or sorted based on the information gain to output a set of candidate control actions for the risk-constrained self-healing joint optimization module to solve.

[0008] Preferably, the belief update module updates the belief according to the following rules: For each fault hypothesis in the set of fault hypotheses, the observation likelihood is calculated based on the observation vector and the predicted observation determined by the fault hypothesis, the current switching state, and the candidate control action. The observation likelihood is minimized within the feasible set of network parameters to obtain the comprehensive likelihood. The comprehensive likelihood is multiplied by the prior probability of the fault hypothesis and normalized to obtain the updated belief.

[0009] Preferably, a fault-related segment is constructed corresponding to each fault hypothesis in the fault hypothesis set. The fault-related segment is determined by the candidate fault branch corresponding to the fault hypothesis and its upstream and downstream segment boundaries. Furthermore, the risk-constrained self-healing joint optimization module and the online safety verification and rollback module jointly apply the following constraint: for any fault hypothesis in the credible fault hypothesis set, it is mandatory that the power grid connectivity caused by the control action does not form an accessible path from any power source node to the fault-related segment corresponding to the fault hypothesis.

[0010] Preferably, the set of credible fault assumptions consists of fault assumptions whose posterior probability is not lower than a preset threshold in the beliefs; and, when solving the control action vector, the risk-constrained self-healing joint optimization module applies voltage boundary constraints and branch current boundary constraints to each fault assumption in the set of credible fault assumptions and the network parameters in the feasible set of network parameters, and together with the reachable path constraints corresponding to the fault-related section, constitutes the conservative achievement of the risk constraints.

[0011] Preferably, the online security verification and rollback module generates rollback control actions according to the following rules: For the first control action in the control action vector, a consistency feasibility check is performed under the set of credible fault assumptions and the set of feasible network parameters. When the check fails, a fallback control action is generated that prohibits the change of switch state and consists only of the change of reactive power setting of distributed power source, the change of reactive power compensation switching and the change of tap changer of voltage regulating transformer. The fallback control action satisfies the reachability path constraint corresponding to the fault-related section.

[0012] Preferably, the risk-constrained self-healing joint optimization module generates the control action vector using a rolling planning method: within each planning cycle, based on the current observation vector, feasible set of network parameters, set of fault hypotheses, and beliefs, a control action vector containing multiple control actions at different times is obtained, and the closed-loop execution module executes only the first control action in the control action vector, and then the multi-source time-series evidence aggregation module updates the observation vector to enter the next planning cycle.

[0013] A preferred method for optimizing fault location and self-healing control in smart grids includes the following steps: S1. Perform time alignment and encoding on continuous measurement data and event-type data, and output the observation vector corresponding to the time. S2. Determine the power grid topology based on the observation vector and switch state information, and output the feasible set of network parameters corresponding to the power grid topology; S3. Based on the observation vector and the power grid topology, generate a set of fault hypotheses including candidate fault branches, fault types and fault resistances; S4. Under the constraints of the feasible set of network parameters, update the belief based on the observation vector and the set of fault hypotheses; S5. Generate a set of candidate control actions based on the belief and calculate the information gain of each candidate control action; S6. The joint objective is formed by the power restoration cost, the action cost, and the belief entropy of the belief, and the control action vector of the rolling plan is obtained from the candidate control action set under the constraints of power grid operation and risk. S7. Before executing the control action vector, the feasibility of the control action vector is checked based on the set of credible fault assumptions and the set of feasible network parameters, and a rollback control action is generated if the check fails. S8. Execute the verified control action and write back new continuous measurement data and event data to update the observation vector and repeat steps S1 to S8.

[0014] Preferably, in step S6, when solving the control action vector, a preset number of fault hypotheses with the highest posterior probability are selected from the fault hypothesis set to form a representative fault hypothesis set, and power grid operation constraints and reachability path constraints corresponding to the fault-related section are simultaneously applied to each fault hypothesis in the representative fault hypothesis set and the network parameters in the network parameter feasible set.

[0015] This invention provides a smart grid fault location and self-healing control optimization system. It has the following beneficial effects: 1. This invention incorporates the control actions of controllable power grid equipment into the fault location decision-making chain. After the control command is issued, the power grid response data generated is fed back to the multi-source time-series evidence aggregation module and the observation vector is reconstructed. This drives the topology-parameter online self-calibration module, the fault hypothesis set generation module, the belief update module, the active detection action planning module, the risk-constrained self-healing joint optimization module, and the online safety verification and rollback module to perform a new round of solution. This forms a closed loop of location-control-observation-relocation, avoiding logical disconnect between the location results and the self-healing control, and ensuring that each control action participates in the evidence update process of fault location confidence convergence.

[0016] 2. Before optimizing the switching action, the present invention first generates a set of credible fault hypotheses by the belief update module, and establishes fault-related section constraints on the set of credible fault hypotheses and the feasible set of network parameters. Then, the same constraint system is reused in the risk-limited self-healing joint optimization module and the online safety verification and rollback module, so that the control action vector maintains a consistent operating safety boundary under multiple fault candidates and parameter boundary conditions, and does not regard fault location as a deterministic input. This ensures that the constraints of the recovery control logic and the fault confidence range are consistent before and after, and avoids constraint conflicts caused by the incorrect inclusion of the power supply path of the candidate fault section into the control scheme.

[0017] 3. This invention performs unified time reference alignment and unambiguous encoding on continuous measurement data and event-type data from different sources and with different sampling rates, constructing observation vectors with constant dimensions and fixed order. This ensures that the input field structure of the feasible set of network parameters, fault hypothesis set, belief, belief entropy, information gain, and candidate control action set in subsequent modules remains unchanged, guaranteeing that the data has a consistent semantic space and index structure when transmitted across time and modules. This provides a unified mathematical and engineering variable description basis for fault location and self-healing control optimization systems. Attached Figure Description

[0018] Figure 1 This is a system structure block diagram of the present invention; Figure 2 This is a schematic diagram of the candidate control action screening logic of the present invention; Figure 3 This is a schematic diagram illustrating the input and output of the belief update in this invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Please see the appendix Figure 1 To be continued Figure 3 This invention provides a smart grid fault location and self-healing control optimization system, comprising: The module includes a multi-source temporal evidence aggregation module, a topology-parameter online self-calibration module, a fault hypothesis set generation module, a belief update module, an active detection action planning module, a risk-constrained self-healing joint optimization module, an online safety verification and rollback module, and a closed-loop execution module. Among them, the multi-source time-series evidence aggregation module outputs observation vectors; The online self-calibration module for topology parameters outputs a feasible set of network parameters; The fault hypothesis set generation module outputs a fault hypothesis set; The belief update module forms beliefs based on observation vectors and fault hypothesis sets under the constraint of feasible set of network parameters; The active detection action planning module generates a set of candidate control actions based on beliefs and calculates information gain. The risk-constrained self-healing joint optimization module uses the power restoration cost, action cost, and belief entropy to form a joint objective, and solves the control action vector of rolling planning from the candidate control action set under the constraints of power grid operation and risk. The online safety verification and rollback module performs a feasibility verification of the control action vector based on the set of credible fault assumptions determined by beliefs and the set of feasible network parameters, and generates a rollback control action when the verification fails. The closed-loop execution module executes the verified control actions and writes back continuous measurement data and event data to update the observation vector and form a closed loop.

[0021] Specifically, the multi-source time-series evidence aggregation module receives continuous measurement data and event-type data, processes the data using a unified time base, and outputs observation vectors. Continuous measurement data includes, but is not limited to, node voltage amplitude, feeder current, node active power, and node reactive power; event-type data includes, but is not limited to, protection action status, switch change status, alarm codes, and communication status. The multi-source time-series evidence aggregation module assigns a unified timestamp field to each data stream and uses a discrete time index t∈N to represent the calculation time of system operation.

[0022] The multi-source time-series evidence aggregation module establishes a fixed-length time alignment window [t] [Δ,t], where Δ>0 is the preset alignment window length. At time t, for each continuous measurement channel, the latest measurement sample within the alignment window is selected as the alignment value for that channel; when there are multiple samples within the alignment window, the arithmetic mean of the samples within the window is used as the alignment value for that channel. For each event-type data channel, the multi-source time-series evidence aggregation module generates an event code within the alignment window. The event code includes at least: an event count, the time difference of the most recent event occurrence, and an event status code. The event count is the number of times the event occurs within the window; the time difference is the difference between t and the time of the most recent event occurrence within the window; the event status code is determined by the event type and the event direction, with the event direction used to distinguish between switch closed and switch open, and protection action and protection reset.

[0023] The multi-source temporal evidence aggregation module assembles the aligned values ​​of continuous measurements into a continuous measurement vector y. t The event codes are composed into an event code vector e. t The observation vector is obtained by concatenating the two in a fixed dimensional order. ; The multi-source time-series evidence convergence module performs a multi-source time-series evidence convergence on the observation vector z. t Perform an integrity check. If a channel has no data within the alignment window, assign a preset missing marker to the corresponding position of that channel and simultaneously add it to the event encoding vector e. t Missing marker bits are written into the data to maintain a constant dimension of the observation vector. The multi-source time-series evidence aggregation module will then process the observation vector z. t The current switch state information is output along with the topology-parameter online self-calibration module, and the observation vector z is also output. tThe output is sent to the fault hypothesis set generation module and the belief update module.

[0024] The online self-calibration module for topology parameters receives observation vectors. In addition to switch status information, the power grid topology is determined and a feasible set of network parameters is output. The power grid topology consists of a set of nodes. Branch set and controllable switch set The connectivity state is determined jointly. For each controllable switch branch... Using binary variables Indicates time The connected states, where This indicates that the branch is connected. This indicates that the branch is disconnected. The online self-calibration module for topology parameters updates the parameters based on the switch position change status and the switch remote signaling status. And based on this, it was determined that... Characterized power grid topology.

[0025] The topology-parameter online self-calibration module defines the network parameter vector. Network parameter vector At least include: each branch road resistance parameters With reactance parameters and load node set Active load of each load node With reactive load The online self-calibration module for topology parameters is based on observation vectors. Continuous measurement vectors in Given a power grid topology, construct a parameter estimation problem and output parameter estimates. In this implementation, the parameter estimation problem is presented in the form of residual minimization: within a preset historical window Internal selection time set Construct the residual vector at sea time generated by the power flow equation and the measurement mapping equation. And solve ; To obtain parameter estimates

[0026] The online self-calibration module for topology parameters constructs a feasible set of network parameters based on the parameter estimation residuals and preset tolerances. In this implementation, the feasible set of network parameters is represented by a weighted quadratic form set: ; in For a pre-defined symmetric positive definite weighting matrix, The set radius is calculated from the parameter estimation residuals. The online self-calibration module for topology parameters outputs the feasible set of network parameters Θt to the belief update module, the risk-constrained self-healing joint optimization module, and the online security verification and rollback module.

[0027] The fault hypothesis set generation module receives observation vectors. Based on the power grid topology, a set of fault hypotheses is generated. The fault hypothesis set generation module first generates a set of candidate faulty branches. Candidate Fault Branch Set The generation is based on the observation vector The protection operation status, switch change status, and continuous measurement changes are constructed as follows: For each feeder, the feeder current change and node voltage change are calculated, and combined with the feeder to which the fault is located and the segment boundary identified by the protection operation, a set of candidate fault branches containing branches within the segment to which the fault belongs is formed.

[0028] The fault hypothesis set generation module predefines a set of fault types. With fault resistor set For any candidate faulty branch For any fault type For any fault resistor Construct a fault hypothesis All failure assumptions are grouped into a failure assumption set. ; The fault hypothesis set generation module generates fault hypothesis sets. Assigning prior probabilities The prior probabilities are generated using a normalized weighting method: for each fault hypothesis... Calculate nonnegative weights , and according to ; Normalization yields the prior probability, where the weights are... The fault hypothesis set is determined by historical fault records of branch lines, equipment status alarms, and statistical weights of fault types. The fault hypothesis set generation module outputs a fault hypothesis set. with prior probability Update the Beliefs module.

[0029] The belief update module updates beliefs according to the following rules: For each fault hypothesis in the fault hypothesis set, the observation likelihood is calculated based on the observation vector and the predicted observation determined by the fault hypothesis, the current switching state, and the candidate control action. The observation likelihood is minimized within the feasible set of network parameters to obtain the comprehensive likelihood. The comprehensive likelihood is multiplied by the prior probability of the fault hypothesis and normalized to obtain the updated belief.

[0030] Specifically, the belief update module receives the observation vector. Fault Hypothesis Set Prior probability Feasible set of network parameters and the current switch status Forming beliefs The belief update module defines the predictive observation operator. in Assuming a failure, For network parameters, For the set of switch states, The control action executed at the previous time step. Predictive observation operator. The calculation is performed according to the following steps: based on the set of switch states Determine the set of conducting branches: based on the control action Update the power supply reactive power settings, reactive power compensation switching status, and tap changer status; in the candidate fault branch Introduction and Fault Type With fault resistor Corresponding fault equivalent branch model; based on power grid topology and network parameters Solve for the power flow variables and map them to the observation vector. The same-dimensional predicted observation vector, the output is

[0031] The belief update module defines the observation noise covariance matrix. And remain unchanged over all times. For any fault assumption With any network parameter The belief update module calculates the observation difference vector. ; and according to ; Calculate the observation likelihood. The belief update module is used in the feasible set of network parameters. Minimize the internal observation likelihood to obtain the comprehensive likelihood. ; The belief update module generates beliefs using normalized Bayesian updates: ; The belief update module further calculates the belief entropy.

[0032] and belief With belief entropy The output is sent to the active detection action planning module and the risk-limited self-healing joint optimization module.

[0033] The control action vector contains control actions at multiple time points, and each control action at each time point is represented by a consistent structure and serves as a component element of the control action vector. Each control action simultaneously includes: changes in the reactive power setting of the distributed power source, changes in the switching of reactive power compensation, changes in the tap changer of the voltage regulating transformer, and changes in the switch state. Furthermore, each candidate control action in the candidate control action set generated by the active detection action planning module satisfies the corresponding action boundary constraints.

[0034] The active detection action planning module determines the information gain according to the following rules: The difference between the belief entropy before the execution of a candidate control action and the expected belief entropy under the predicted observation distribution after the execution of the candidate control action is used as the information gain of the candidate control action. The candidate control action set is then filtered or sorted based on the information gain to output a set of candidate control actions for the risk-constrained self-healing joint optimization module.

[0035] Specifically, the active detection and incentive planning module receives beliefs. Belief Entropy Feasible set of power grid topology and network parameters and the current switch state set The system generates a set of candidate control actions and calculates the information gain. The active detection action planning module defines a unified structure for the control actions, which are: ; in Define a change vector for reactive power of distributed generation. This is the reactive power compensation switching change vector. For the tap change of the voltage regulating transformer, This represents the vector of switch state changes. The active detection action planning module generates a set of candidate control actions based on preset action boundary constraints. The action boundary constraints include at least the following: the upper and lower bounds of reactive power setting changes for each distributed power source, the set of allowable taps for reactive power compensation switching changes, the allowable integer range of tap changes for voltage regulating transformers, and the set of allowable values ​​for switch state changes.

[0036] The active detection action planning module is based on belief. Construct a set of credible failure hypotheses. The set of credible failure hypotheses is defined as follows: ; in The preset threshold is used. The active detection action planning module assumes several faults. Construct fault-related sections Fault-related section From candidate faulty branch And the upstream and downstream segmentation ranges defined by the sectionalizing switch, and the fault-related sections. The corresponding set of branches can be mapped to the corresponding set of nodes. .

[0037] The active detection action planning module detects the set of candidate control actions. Perform reachability path constraint screening. For any candidate control action... The change in the switching state of the candidate control action Obtain the set of switch states after execution In the context of power grid topology and switch state set Calculate the set of power nodes in the connected graph. The set of reachable nodes When for all All meet ; If the above conditions are met, retain the candidate control action; if the above conditions are not met, remove the candidate control action. The resulting set of candidate control actions is obtained after filtering.

[0038] The active detection action planning module calculates the information gain for each candidate control action after screening. Define the expected belief entropy under the predicted observation distribution as: And define information gain as ; For calculation The active detection and incentive planning module constructs a finite set of predictive observation samples. For each fault hypothesis Network parameter estimates With the set of switch states after execution Calculate and predict observations And according to the preset observation perturbation set Generate samples ; For each sample Calculate the corresponding updated beliefs according to the rules of the belief update module. And calculate The active detection action planning module uses b t (h) represents the weighted average of the sample entropy to obtain E[S(b)]. t+1 The active detection action planning module outputs the information gain of each candidate control action and outputs the set of candidate control actions for the risk-constrained self-healing joint optimization module to solve.

[0039] The fault-related segment is constructed for each fault hypothesis in the set of fault hypotheses. The fault-related segment is determined by the candidate fault branch corresponding to the fault hypothesis and its upstream and downstream segment boundaries. Furthermore, the risk-constrained self-healing joint optimization module and the online safety verification and rollback module jointly apply the following constraints: for any fault hypothesis in the set of credible fault hypotheses, it is mandatory to satisfy that the power grid connectivity relationship caused by the control action does not form an accessible path from any power node to the fault-related segment corresponding to the fault hypothesis.

[0040] The set of credible fault assumptions consists of fault assumptions whose posterior probability is not lower than a preset threshold. Furthermore, when solving the control action vector, the risk-constrained self-healing joint optimization module applies voltage boundary constraints and branch current boundary constraints to each fault assumption in the set of credible fault assumptions and the network parameters in the feasible set of network parameters. Together with the reachable path constraints corresponding to the fault-related section, they constitute a conservative achievement of the risk constraints.

[0041] The risk-constrained self-healing joint optimization module uses a rolling planning approach to generate control action vectors: within each planning cycle, based on the current observation vector, feasible set of network parameters, set of fault hypotheses, and beliefs, it solves for control action vectors containing control actions at multiple time points. The closed-loop execution module executes only the first control action in the control action vector, and then the multi-source time-series evidence aggregation module updates the observation vector to enter the next planning cycle.

[0042] Specifically, the risk-limited self-healing joint optimization module receives observation vectors. Feasible set of network parameters Fault Hypothesis Set ,belief Belief Entropy Candidate control action set Information gain and the current switch state set The control action vector of the rolling program is obtained under the constraints of power grid operation and risk; the risk-constrained self-healing joint optimization module sets the length of the rolling program. And define the control action vector as ; Each of them All are controlled by a unified structure Risk-limited self-healing joint optimization module introduces power supply state variables Indicates load node At any moment The power supply status, among which This indicates that the load node is being powered. This indicates that the load node is not receiving power. The risk-constrained self-healing joint optimization module constructs the joint objective function.

[0043] in Preset weights At the cost of restoring power, This represents the cost of the operation. In this implementation, the cost of power restoration is expressed as a weighted, unpowered load:

[0044] in Weight of load nodes This refers to the active load at the load node. In this implementation, the action cost is a weighted sum of the action amplitude and the number of switch changes:

[0045] in To preset non-negative weighting constants, This is a set of reactive power compensation devices. To ensure that the objective function is computable with respect to belief evolution during the solution process, the risk-constrained self-healing joint optimization module employs the equivalent substitution of information gain: for each time step within the planning period... by Express the belief entropy term, where The active detection action planning module calculates the parameters according to the aforementioned rules and inputs them as known parameters.

[0046] The risk-constrained self-healing joint optimization module applies grid operation constraints. In this implementation, these constraints are represented by branch power flow variables and voltage variables. For the branch set... Any of the conducting branches Introducing branch line active current Branch road without power flow Branch current square and the square of the node voltage Power grid operation constraints include at least power balance constraints and boundary constraints: power balance constraints apply to each node. Defined according to the branch direction, satisfying

[0047]

[0048] in For nodes The generation injection is zero for non-power nodes. Voltage boundary constraints are... ; Branch current boundary constraints are ; For controllable switch branches The risk-limited self-healing joint optimization module will switch on / off states. Associated with the current variable, so that when Time branch path power flow variables satisfy .

[0049] The risk-constrained self-healing joint optimization module imposes risk constraints and achieves them simultaneously according to the set of credible failure assumptions and the feasible set of network parameters. The risk-constrained self-healing joint optimization module is based on beliefs... Constructing a set of credible fault hypotheses And for each planning moment Use and The same construction rules are obtained The risk constraint consists of the following set of constraints: for all And for all All conditions are forcibly met, including voltage boundary constraints and branch current boundary constraints within the power grid operation constraints. To reduce the solution scale, the risk-constrained self-healing joint optimization module further constructs a representative set of fault assumptions. Representative Fault Hypothesis Set From the set of fault assumptions The hypothesis consists of a preset number of fault hypotheses with the highest posterior probability, and is based on... They participate in the application of constraints in a certain way.

[0050] The risk-constrained self-healing joint optimization module applies reachability path constraints corresponding to fault-related segments. For any... Construct fault-related sections And obtain the corresponding node set For each plan, heat it up. , from the set of switch states Generate a connected graph and calculate the set of power nodes. The set of reachable nodes Risk-limited self-healing joint optimization module mandatory satisfaction ; This is to ensure that the grid connectivity does not create a reachable path from any power source node to the fault-related section.

[0051] Under the aforementioned joint objective function, power grid operation constraints, and risk constraints, the risk-constrained self-healing joint optimization module optimizes the control action vector A. t Solve and output the control action vector. The risk-constrained self-healing joint optimization module outputs the control action vector to the online safety verification and rollback module.

[0052] The online security verification and rollback module generates rollback control actions according to the following rules: For the first control action in the control action vector, a consistency feasibility check is performed under the set of credible fault assumptions and the set of feasible network parameters. When the check fails, a backoff control action is generated that prohibits the change of switch state and consists only of the change of reactive power setting of distributed power source, the change of reactive power compensation switching and the change of tap changer of voltage regulating transformer. The backoff control action satisfies the reachability path constraint corresponding to the fault-related section.

[0053] Specifically, the online security verification and rollback module receives the control action vector, the set of credible fault hypotheses, and the set of feasible network parameters. It performs a feasibility check on the control action vector and generates a rollback control action if the check fails. The online security verification and rollback module selects the first control action from the control action vector. As the verification object, construct the set of switch states after executing the control action. And control settings updated equipment status. The online safety verification and rollback module performs consistency verification according to the same grid operation constraints and reachability path constraints as the risk-constrained self-healing joint optimization module: for all With all Verify whether the voltage boundary constraints, branch current boundary constraints, and reachability path constraints corresponding to the fault-related sections are valid.

[0054] When the verification is successful, the online safety verification and rollback module outputs the verified control action to the closed-loop execution module. When the verification fails, the online safety verification and rollback module generates a rollback control action. The structure of the rollback control action is the same as the control action, denoted as... ; in All values ​​are set to zero, thus preventing the backoff control action from changing its switch state; and the action boundary constraints are applied to... The amplitude is trimmed to meet the corresponding action boundary constraints. The online safety verification and rollback module re-verifies the rollback control action according to the above consistency verification process, and outputs the rollback control action that passes the verification to the closed-loop execution module.

[0055] The closed-loop execution module receives verified control actions or rollback control actions and sends them to the corresponding execution devices. The module generates control commands for four types of changes in control actions: a reactive power setting update command for changes in distributed power source reactive power settings; a switching position update command for changes in reactive power compensation switching; a tap position update command for changes in tap changers of voltage regulating transformers; and a closing / opening control command for changes in switch status. The closed-loop execution module sends the control commands to the execution devices via a preset communication interface and receives execution confirmation information from the execution devices. This confirmation information includes at least the command number, execution result status, and execution completion timestamp.

[0056] After the closed-loop execution module completes the control action, it collects continuous measurement data and event data corresponding to that control action and writes the collected data back to the multi-source time-series evidence aggregation module. Upon receiving the written-back data, the multi-source time-series evidence aggregation module generates a new observation vector z according to the aforementioned time alignment and encoding rules. t+1 The risk-limited self-healing joint optimization module uses a rolling planning approach to resolve the control action vector in the next planning cycle using only the new observation vector and the updated feasible set of network parameters, fault hypothesis set, and beliefs. The online safety verification and rollback module performs a feasibility verification on the first control action of the new control action vector, thus forming a closed-loop operation process.

[0057] The optimization method for fault location and self-healing control in smart grids includes the following steps: S1. Perform time alignment and encoding on continuous measurement data and event-type data, and output the observation vector corresponding to the time. S2. Determine the power grid topology based on the observation vector and switch state information, and output the feasible set of network parameters corresponding to the power grid topology; S3. Generate a set of fault hypotheses, including candidate fault branches, fault types, and fault resistances, based on observation vectors and power grid topology. S4. Under the constraint of the feasible set of network parameters, update the belief based on the observation vector and the set of fault hypotheses; S5. Generate a set of candidate control actions based on beliefs and calculate the information gain of each candidate control action; S6. The joint objective is formed by the power restoration cost, action cost, and belief entropy, and the control action vector of rolling planning is obtained from the candidate control action set under the constraints of power grid operation and risk. S7. Before executing the control action vector, the feasibility of the control action vector is checked based on the set of credible fault assumptions and the set of feasible network parameters, and a backoff control action is generated if the check fails. S8. Execute the verified control action and write back the new continuous measurement data and event data to update the observation vector and repeat steps S1 to S8.

[0058] In step S6, when solving for the control action vector, a preset number of fault hypotheses with the highest posterior probability are selected from the fault hypothesis set to form a representative fault hypothesis set. Power grid operation constraints and reachability path constraints corresponding to the fault-related sections are simultaneously applied to each fault hypothesis in the representative fault hypothesis set and the network parameters in the network parameter feasible set.

[0059] Specifically, when step S1 is executed, the continuous measurement data is aligned according to a preset time window, and the alignment value of each measurement channel within the time window is extracted to form a continuous measurement vector; the event data is converted into an event encoding vector according to a preset encoding rule, and the event encoding vector includes at least the event count, the time difference of the most recent event, and the event status code; the continuous measurement vector and the event encoding vector are concatenated in a fixed dimension order to obtain the observation vector.

[0060] When step S2 is executed, the switch state set is updated and the power grid topology is determined based on the switch change state and switch remote signaling state; the network parameter estimate is calculated based on the continuous measurement components in the observation vector and the power grid topology, and a feasible set of network parameters is generated according to the preset set construction rules.

[0061] When step S3 is executed, a set of candidate fault branches is generated based on the protection action status, switch change status and continuous measurement change information in the observation vector; for each candidate fault branch in the set of candidate fault branches, a set of fault hypotheses is generated by enumerating and combining them according to the preset set of fault types and the preset set of fault resistors; and a prior probability is generated for each fault hypothesis in the set of fault hypotheses.

[0062] When step S4 is executed, for each fault hypothesis in the fault hypothesis set, under the equipment setting conditions determined by the current switch state and the control action executed at the previous time, predictive observations are generated based on the network parameters in the feasible set of network parameters; the predictive observations and observation vectors are subjected to consistency measurement to obtain the observation likelihood; the observation likelihoods are aggregated within the feasible set of network parameters to obtain the comprehensive likelihood; the comprehensive likelihood and prior probability are normalized and updated to obtain the belief, and the belief entropy is calculated.

[0063] When step S5 is executed, a set of candidate control actions is generated according to the unified structure of control actions. The unified structure of control actions includes changes in reactive power setting of distributed power sources, changes in reactive power compensation switching, changes in tap change of voltage regulating transformer, and changes in switch status. The set of candidate control actions is generated and filtered according to action boundary constraints. A set of credible fault hypotheses is determined by belief. A fault-related segment is constructed for each fault hypothesis in the set of credible fault hypotheses. The reachability path constraints corresponding to the fault-related segments are applied to the set of candidate control actions, and candidate control actions that do not meet the constraints are eliminated. For the remaining candidate control actions, the information gain is calculated based on the belief entropy before the execution of the candidate control action and the expected belief entropy after the execution of the candidate control action, and the information gain of each candidate control action is output.

[0064] When step S6 is executed, a control action vector containing multiple control actions at different times is generated under a preset rolling planning length. Each control action in the control action vector is represented according to the unified structure of the control action. During the solution process, voltage boundary constraints and branch current boundary constraints are applied to each fault hypothesis in the set of credible fault hypotheses and the network parameters in the set of feasible network parameters. Together with the reachable path constraints corresponding to the fault-related section, they constitute a conservative implementation of the risk constraints.

[0065] When step S7 is executed, the first control action in the control action vector is checked for consistency feasibility. The consistency feasibility check is consistent with the power grid operation constraints and risk constraints in step S6. When the consistency feasibility check fails, a rollback control action is generated. The rollback control action prohibits changes in switch status and only includes changes in reactive power settings of distributed power sources, changes in reactive power compensation switching, and changes in tap changers of voltage regulating transformers, and satisfies the reachability constraints of the fault-related section.

[0066] When step S8 is executed, control commands corresponding to the control actions are generated and issued, and execution confirmation information is received; after the control actions are completed, continuous measurement data and event data corresponding to the control actions are collected and written back to trigger the next round of observation vector generation and subsequent steps.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart grid fault location and self-healing control optimization system, characterized in that, include: The module includes a multi-source temporal evidence aggregation module, a topology-parameter online self-calibration module, a fault hypothesis set generation module, a belief update module, an active detection action planning module, a risk-constrained self-healing joint optimization module, an online safety verification and rollback module, and a closed-loop execution module. The multi-source time-series evidence aggregation module outputs an observation vector. The online self-calibration module for topology parameters outputs a feasible set of network parameters. The fault hypothesis set generation module outputs a fault hypothesis set; The belief update module forms beliefs based on the observation vector and the fault hypothesis set under the constraints of the feasible set of network parameters; The active detection action planning module generates a set of candidate control actions based on the belief and calculates the information gain. The risk-constrained self-healing joint optimization module uses the power restoration cost, action cost, and belief entropy of the belief to form a joint objective, and solves the control action vector of the rolling plan from the candidate control action set under the constraints of power grid operation and risk. The online security verification and rollback module performs a feasibility verification of the control action vector based on the set of credible fault assumptions determined by the belief and the set of feasible network parameters, and generates a rollback control action when the verification fails. The closed-loop execution module executes the verified control actions and writes back continuous measurement data and event data to update the observation vector to form a closed loop.

2. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The control action vector includes control actions at multiple times, and the control actions at each time are represented by a consistent structure and serve as the constituent elements of the control action vector; wherein, each control action simultaneously includes: changes in the reactive power setting of the distributed power source, changes in the switching of reactive power compensation, changes in the tap changer of the voltage regulating transformer, and changes in the switch state; and, each candidate control action in the candidate control action set generated by the active detection action planning module satisfies the corresponding action boundary constraints.

3. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The active detection action planning module determines the information gain according to the following rules: The difference between the belief entropy before the execution of the candidate control action and the expected belief entropy under the predicted observation distribution after the execution of the candidate control action is used as the information gain of the candidate control action. The candidate control action set is then filtered or sorted based on the information gain to output a set of candidate control actions for the risk-constrained self-healing joint optimization module to solve.

4. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The belief update module updates the belief according to the following rules: For each fault hypothesis in the set of fault hypotheses, the observation likelihood is calculated based on the observation vector and the predicted observations determined by the fault hypothesis, the current switching state, and the candidate control action; the minimum value of the observation likelihood is taken within the feasible set of network parameters to obtain the comprehensive likelihood; The updated belief is obtained by multiplying the combined likelihood with the prior probability of the fault hypothesis and normalizing the result.

5. The smart grid fault location and self-healing control optimization system according to claim 4, characterized in that, Construct a fault-related segment corresponding to each fault hypothesis in the fault hypothesis set. The fault-related segment is determined by the candidate fault branch corresponding to the fault hypothesis and its upstream and downstream segment boundaries. Furthermore, the risk-constrained self-healing joint optimization module and the online safety verification and rollback module jointly apply the following constraint: for any fault hypothesis in the credible fault hypothesis set, it is mandatory that the power grid connectivity relationship caused by the control action does not form an reachable path from any power source node to the fault-related segment corresponding to the fault hypothesis.

6. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The set of credible fault assumptions consists of fault assumptions whose posterior probability is not lower than a preset threshold in the beliefs; and, when solving the control action vector, the risk-constrained self-healing joint optimization module applies voltage boundary constraints and branch current boundary constraints to each fault assumption in the set of credible fault assumptions and the network parameters in the feasible set of network parameters, and together with the reachable path constraints corresponding to the fault-related section, constitutes the conservative achievement of the risk constraints.

7. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The online security verification and rollback module generates rollback control actions according to the following rules: For the first control action in the control action vector, a consistency feasibility check is performed under the set of credible fault assumptions and the set of feasible network parameters. When the check fails, a fallback control action is generated that prohibits the change of switch state and consists only of the change of reactive power setting of distributed power source, the change of reactive power compensation switching and the change of tap changer of voltage regulating transformer. The fallback control action satisfies the reachability path constraint corresponding to the fault-related section.

8. The smart grid fault location and self-healing control optimization system according to claim 1, characterized in that, The risk-constrained self-healing joint optimization module generates the control action vector using a rolling planning approach: within each planning cycle, based on the current observation vector, feasible set of network parameters, set of fault hypotheses, and beliefs, it solves for a control action vector containing multiple control actions at different times. The closed-loop execution module then executes only the first control action in the control action vector, and subsequently, the multi-source time-series evidence aggregation module updates the observation vector to enter the next planning cycle.

9. A method for optimizing fault location and self-healing control in smart grids, characterized in that, The smart grid fault location and self-healing control optimization system according to any one of claims 1-8 includes the following steps: S1. Perform time alignment and encoding on continuous measurement data and event-type data, and output the observation vector corresponding to the time. S2. Determine the power grid topology based on the observation vector and switch state information, and output the feasible set of network parameters corresponding to the power grid topology; S3. Based on the observation vector and the power grid topology, generate a set of fault hypotheses including candidate fault branches, fault types and fault resistances; S4. Under the constraints of the feasible set of network parameters, update the belief based on the observation vector and the set of fault hypotheses; S5. Generate a set of candidate control actions based on the belief and calculate the information gain of each candidate control action; S6. The joint objective is formed by the power restoration cost, the action cost, and the belief entropy of the belief, and the control action vector of the rolling plan is obtained from the candidate control action set under the constraints of power grid operation and risk. S7. Before executing the control action vector, the feasibility of the control action vector is checked based on the set of credible fault assumptions and the set of feasible network parameters, and a rollback control action is generated if the check fails. S8. Execute the verified control action and write back new continuous measurement data and event data to update the observation vector and repeat steps S1 to S8.

10. The smart grid fault location and self-healing control optimization method according to claim 9, characterized in that, In step S6, when solving the control action vector, a preset number of fault hypotheses with the highest posterior probability are selected from the fault hypothesis set to form a representative fault hypothesis set. Power grid operation constraints and reachability path constraints corresponding to the fault-related section are simultaneously applied to each fault hypothesis in the representative fault hypothesis set and the network parameters in the network parameter feasible set.

Images