An adaptive relay protection setting value on-line scheduling system containing distributed power supply

Abstract

The application relates to the technical field of power system protection and control, and discloses an adaptive relay protection setting value online scheduling system containing a distributed power supply. The system is aimed at the mixed access characteristics of the inverter interface type distributed power supply, the synchronous generator type distributed power supply, the asynchronous generator type distributed power supply and the energy storage type distributed power supply in a distribution network. The system realizes online identification of the dynamic fault characteristics of the distributed power supply by real-time collection of multi-source heterogeneous data of PMU, SCADA, distributed power supply operation states, protection device information and meteorological data, and generates a protection control integrated scheduling strategy in combination with a deep reinforcement learning algorithm. The application can dynamically adjust various parameters of differential protection, and realizes accurate fault positioning by using distributed optical fiber sensing, traveling wave distance measurement and current waveform analysis, thereby solving the problem of insufficient adaptability of the traditional fixed setting value mode under high proportion of distributed power supply access.

Description

Technical Field

[0001] This invention relates to the field of power system protection and control technology, and more specifically, to an adaptive relay protection setting online dispatching system containing distributed power sources. Background Technology

[0002] Currently, with the increasing penetration of distributed generation (DG) in distribution networks, especially the mixed access of various types of distributed generation (DG), such as inverter-interfaced distributed generation (IIDG) and synchronous generator-based distributed generation (SGDG), the operating characteristics and fault behavior of distribution networks exhibit significant dynamism and complexity.

[0003] Traditional relay protection setting methods typically involve offline calculations and presets based on the most severe fault conditions or typical operating modes of the system. These fixed or preset protection settings are ill-suited to the challenges posed by distributed power source output fluctuations, bidirectional power flow, and potential dynamic changes in network topology. This can easily lead to protection devices failing to operate (failing to correctly identify and isolate the fault) or malfunctioning (operating incorrectly in normal or non-faulty areas) under certain unexpected operating conditions, thereby affecting power supply reliability and system safety and stability.

[0004] Furthermore, existing online scheduling methods for relay protection settings and active management measures of the distribution network (such as distributed generation output control, energy storage system scheduling, and flexible load response) are often independent of each other, lacking an effective collaborative optimization mechanism. This limits the full utilization of the potential of distributed generation and makes it difficult to achieve globally optimal operation of the entire distribution system under safety constraints. At the same time, the integration of hybrid distributed generation alters the characteristics of short-circuit current (for example, inverter-interface type distributed generation typically exhibits current-limiting characteristics, while synchronous generator type distributed generation provides larger short-circuit currents). This increases the difficulty of accurately identifying the dynamic contribution of each distributed generation to the fault online and precisely locating the fault point within the protection area (especially in differential protection areas).

[0005] Therefore, existing online scheduling methods are insufficient in terms of intelligence, adaptability, and deep integration with the overall operation strategy of the distribution network, which are technical problems that urgently need to be solved in related technical fields. Summary of the Invention

[0006] This invention provides an online scheduling system for adaptive relay protection settings with distributed power sources, which solves the technical problems in related technologies such as insufficient adaptability of traditional relay protection settings in distribution networks with high proportions and multiple types of distributed power sources, disconnect between protection strategies and control measures, difficulty in fault identification and location, and poor performance of the overall scheduling scheme in terms of intelligence and adaptability.

[0007] This invention provides an adaptive relay protection setting online scheduling system with distributed power sources, comprising:

[0008] The data acquisition module is used to collect multi-source heterogeneous data from the power distribution network in real time and to build a panoramic view of the power distribution network operation.

[0009] The fault characteristic identification module is used to identify the dynamic fault characteristic parameters of hybrid distributed power sources in the distribution network online.

[0010] The collaborative scheduling strategy generation module, based on a deep reinforcement learning algorithm, generates an integrated collaborative scheduling strategy for protection and control according to the panoramic operation view and dynamic fault characteristic parameters. The collaborative scheduling strategy includes protection setting adjustment instructions and control instructions for active distribution network management measures.

[0011] The adaptive setting and fault location module is used to adaptively set the online settings of specific protection based on dynamic fault characteristic parameters and collaborative scheduling strategies, and to achieve accurate location of the fault area by combining high-frequency sampling data analysis technology.

[0012] The strategy execution and optimization module is used to execute collaborative scheduling strategies online and to verify and continuously optimize the strategies.

[0013] In a preferred embodiment, the multi-source heterogeneous data collected by the data acquisition module includes:

[0014] At least one of the following: synchronous phasor measurement unit data, SCADA system data, operating status parameters of various distributed power sources, protection device information, and meteorological data.

[0015] In a preferred embodiment, the fault characteristic identification module uses parameter identification algorithms or machine learning techniques to estimate online the equivalent circuit parameters or fault current response characteristics of at least one of the inverter interface type distributed power source and the synchronous generator type distributed power source.

[0016] In a preferred embodiment, the cooperative scheduling strategy generation module includes:

[0017] The problem of distribution network operation and protection setting scheduling is constructed as a Markov decision process, wherein the state input includes at least one of real-time grid measurement, distributed generation status and current protection setting, the action output includes at least one of protection setting adjustment command and control command of active distribution network management measures, and the reward function comprehensively considers at least one of distribution network operation economy, distributed generation absorption capacity, system safety margin and penalty for improper protection behavior.

[0018] A deep reinforcement learning algorithm is used to train the agent to generate a cooperative scheduling strategy.

[0019] In a preferred embodiment, the adaptive setting and fault location module dynamically adjusts at least one of the following for line differential protection or transformer differential protection: operating current threshold, ratio braking coefficient, or vector compensation parameter.

[0020] In a preferred embodiment, the adaptive setting and fault location module employs traveling wave ranging technology based on distributed optical fiber sensing, or analyzes the minute differences in the arrival times of the current waveforms on both sides of the differential protection unit to achieve accurate location of the fault point within the protection area.

[0021] In a preferred embodiment, the strategy execution and optimization module verifies the effectiveness and security of the collaborative scheduling strategy using a digital twin simulation environment or a small-scale pilot application;

[0022] Furthermore, by leveraging accumulated operational data and scheduling experience, the deep reinforcement learning algorithm model is periodically or event-triggered for retraining and iterative optimization.

[0023] In a preferred embodiment, the deep reinforcement learning algorithm is an Actor-Critic type algorithm.

[0024] In a preferred embodiment, the control commands for active distribution network management measures include at least one of distributed generation active or reactive power regulation commands and energy storage charging and discharging commands.

[0025] In a preferred embodiment, a computer-readable storage medium is provided for storing computer-readable instructions that, when read by a computer, enable the operation of an online scheduling system for adaptive relay protection settings containing distributed power sources.

[0026] The beneficial effects of this invention are as follows:

[0027] This invention achieves online adaptive adjustment of relay protection settings and accurate and rapid location of fault areas by sensing the dynamic operating status of the distribution network in real time and identifying the fault characteristics of hybrid distributed power sources online. It also innovatively uses deep reinforcement learning algorithms to construct an integrated collaborative scheduling strategy for protection and control. At the same time, it ensures that the scheduling strategy can be continuously iterated and optimized based on actual operating data.

[0028] Compared with existing technologies, this invention significantly solves the problems of poor adaptability, reduced selectivity, and easy failure to operate or maloperate in complex distribution networks with high proportions and multiple types of distributed generation (DG) under traditional fixed setting protection; overcomes the defects of the separation between protection setting adjustment and distribution network active management measures, which fails to give full play to the supporting role of DG in the system; improves the situation of variable fault characteristics caused by DG access and insufficient accuracy of traditional fault identification and location methods; and enhances the intelligence and adaptability of dispatching decisions by introducing artificial intelligence technology.

[0029] This invention achieves beneficial technical effects such as improving the accuracy, selectivity, and speed of relay protection actions; optimizing the overall economic efficiency and reliability of distribution network operation (e.g., improving DG absorption capacity and reducing network losses); enhancing rapid fault response and precise handling capabilities; improving the intelligence, automation, and refinement of distribution network dispatching; and comprehensively ensuring the safe and stable operation of the distribution system under high-penetration DG access. Attached Figure Description

[0030] Figure 1 This is a block diagram of an adaptive relay protection setting online scheduling system with distributed power supply according to the present invention. Detailed Implementation

[0031] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0032] At least one embodiment of the present invention discloses an adaptive relay protection setting online scheduling system with distributed power sources, such as... Figure 1 As shown, it includes:

[0033] The data acquisition module is used to collect multi-source heterogeneous data from the power distribution network in real time and to build a panoramic view of the power distribution network operation.

[0034] Specifically, the following steps are included:

[0035] Step 1.1: Acquire data from the synchronous phasor measurement unit (PMU);

[0036] Acquire high-frequency real-time data such as voltage, current phasors, frequency, and rate of change of frequency at key nodes in the distribution network.

[0037] Step 1.2: Collect SCADA system data;

[0038] Acquire routine operating data such as the status of circuit breakers, disconnectors, transformer loads, and line power flow in the distribution network.

[0039] Step 1.3: Collect operating status parameters of distributed power sources;

[0040] For different types of DG, information such as active power, reactive power, voltage, current, and internal operating parameters (such as DC side voltage and energy storage status of IIDG) at their grid connection points are collected.

[0041] Step 1.4 Collect information on protection devices and meteorological data: Collect information on the type, settings, and operation of existing protection devices, as well as meteorological data related to DG output (such as light intensity, wind speed, etc.).

[0042] Step 1.5: Construct a dynamic panoramic view of the power distribution network operation;

[0043] The collected heterogeneous data from multiple sources are fused, processed, and stored to form a panoramic view that reflects the real-time operating status of the power distribution network, providing a unified data interface for subsequent analysis.

[0044] The fault characteristic identification module is used to identify the dynamic fault characteristic parameters of hybrid distributed power sources in the distribution network online.

[0045] Specifically, the following steps are included:

[0046] Step 2.1, DG fault characteristic estimation based on parameter identification algorithm;

[0047] Using high-frequency real-time data (such as voltage and current waveforms before and after a fault in PMU data), and employing parameter identification algorithms such as least squares method, Kalman filtering, or particle swarm optimization, the equivalent Thevenin impedance of IIDG and SGDG is estimated online.

[0048] Z DG，eq =R DG，eq +jX DG，eq ;

[0049] Among them, Z DG，eq R represents the equivalent internal impedance of DG. DG，eq X is the equivalent resistance. DG，eqFor equivalent reactance, j represents the weighting coefficient of equivalent reactance, and eq stands for "equivalent", referring to an equivalent model that simplifies the complex internal circuitry of a distributed power source into a single impedance.

[0050] In some implementations, other advanced system identification techniques, such as recursive least squares or extended Kalman filtering, can also be used to adapt to different data characteristics and computational resource constraints.

[0051] Step 2.2, DG fault characteristic prediction based on machine learning;

[0052] Construct machine learning models (e.g., Recurrent Neural Networks (RNNs), Long Short-Term Memory Networks (LSTMs), or Gated Recurrent Units (GRUs)) and input historical fault data and real-time operational data to predict the dynamic response characteristics of different types of distributed generation (DG) under different fault types (e.g., single-phase grounding, phase-to-phase short circuits) and fault locations, including the magnitude of the fault current I. fault_DG Phase angle φ fault_DG and the duration T of fault current injection inject .

[0053] Step 2.3: Output the dynamic fault characteristic parameters of the DG;

[0054] The identified or predicted dynamic fault characteristic parameters of each DG (such as Z) DG，eq I DG，N φ fault_DG φ fault_DG T inject The data is integrated to form structured data, which is then used for subsequent protection scheduling strategy generation and setting.

[0055] The collaborative scheduling strategy generation module, based on a deep reinforcement learning algorithm, generates an integrated collaborative scheduling strategy for protection and control according to the panoramic operation view and dynamic fault characteristic parameters. The collaborative scheduling strategy includes protection setting adjustment instructions and control instructions for active distribution network management measures.

[0056] Specifically, the following steps are included:

[0057] Step 3.1, construct the Markov decision process model;

[0058] The problem of coordinated scheduling of protection and control in power distribution networks is formalized as a Markov Decision Process (MDP) model.

[0059] The state space S is defined to include the overall state, such as line power flow, node voltage, and DG output P. DG Q DG Current protection settings, etc.;

[0060] Operating space A includes instructions for adjusting protection settings (e.g., for differential protection, adjusting the operating current threshold I). op_thresh Ratio braking slope k slope ) and control commands for active management measures (such as DG active power regulation ΔP) DG Reactive power regulation ΔQ DG Energy storage charging and discharging power P ESS );

[0061] Reward function R t The design comprehensively reflects the system's operational economy, safety, DG absorption rate, and the correctness of protection actions. For example:

[0062] R t =w1·E econ -w2·P loss +w3·E DG_util -w4·P penalty_prot ;

[0063] Among them, R t Let E be the reward function at time t, used to evaluate the merits of taking a specific action in the current state; econ The economic benefits of economic scheduling represent the economic efficiency of system operation; P loss Network loss refers to the active power loss in the power system; E DG_util DG absorption capacity represents the amount of distributed generation successfully absorbed by the system; P penalty_prot The penalty for misconduct is used to quantify the negative impact of malfunction or failure of protection devices; w1, w2, w3, and w4 represent the weight coefficients of economic dispatch revenue, network loss, DG consumption, and penalty for misconduct, respectively.

[0064] In some implementations, the reward function may also include penalties for voltage exceeding limits, penalties for the magnitude of control actions (to reduce unnecessary frequent actions), etc., to more comprehensively guide the learning direction of the agent.

[0065] Step 3.2, train the DRL agent;

[0066] The agent is trained using an Actor-Critic type DRL algorithm. Specifically, both the Actor network and the Critic network can be constructed using deep neural networks.

[0067] Actor network π(s) t ;θ π Accept the current state s t As input, the output is a defined action a. t (In DDPG) or the probability distribution of actions (in stochastic policy algorithms such as SAC), where st The system state at time t represents the current state of the distribution network, including information such as the operating state of the distributed generation system and the current protection settings; θ π For the Actor network, the parameter set includes the network weights and biases, π(s) t ;θ π Let θ be the policy function, representing the condition where the parameter is θ. π Under the condition of input state s t The action strategy output afterward.

[0068] Critic network Q(s) t ,a t ;θ Q Input the current state s t and the corresponding action a t Output the value estimate Q of the state-action pair, where s t The system state at time t represents the current state of the distribution network, including information such as the operating state of the distributed generation system and the current protection settings; θ Q This represents the set of parameters of the Critic network, including the network's weights and biases, Q(s). t a t ;θ Q Let θ be the action value function, representing the action value function with parameter θ. Q Under the condition, state s t Take action a t The long-term cumulative return estimate.

[0069] Step 3.3: Output the collaborative scheduling strategy;

[0070] The trained DRL agent (i.e. its Actor network) is the desired integrated protection and control collaborative scheduling strategy model.

[0071] This model can adapt to the real-time input of the distribution network status s t The system outputs optimal protection setting adjustment commands and active management control commands online. t .

[0072] The adaptive setting and fault location module is used to adaptively set the online settings of specific protection based on dynamic fault characteristic parameters and collaborative scheduling strategies, and to achieve accurate location of the fault area by combining high-frequency sampling data analysis technology.

[0073] Specifically, the following steps are included:

[0074] Step 4.1, the differential protection setting is adjusted adaptively;

[0075] Taking line differential protection as an example, based on the dynamic fault current contribution characteristics (such as current limiting amplitude and injection time) of the DG connected at both ends of the line under different operating modes, and the setting adjustment direction given by the DRL collaborative scheduling strategy, the operating current threshold I of the differential protection is adjusted online. op_thresh and ratio braking coefficient k slope .

[0076] For example, when a certain IIDG is identified as exhibiting significant current-limiting characteristics in the early stages of a fault, the Id can be appropriately reduced. op_thresh To improve sensitivity, or adjust k slope The inflection point is determined to accommodate its nonlinear fault current characteristics.

[0077] In some implementations, the adjustment of the setpoint can also be determined by a combination of factors such as the load level of the line, the estimated fault type, or the system operation mode (such as grid connection or islanding operation) to achieve more refined protection coordination.

[0078] Step 4.2, enhanced precise location of the fault area;

[0079] When the differential protection has been activated or a fault is suspected in the protection zone, high-frequency PMU data or data from a dedicated distributed fiber optic sensing system can be used to accurately locate the fault.

[0080] For example, an algorithm based on traveling wave ranging can be used to analyze the time difference Δt between the high-frequency traveling wave signal generated at the moment of the fault and the PMU or fiber optic sensing unit at both ends of the line. arrival The distance to the fault point is calculated by combining the line parameters.

[0081] The calculation formula can be simplified to:

[0082] L f =(L line -v wave ·Δt arrival ) / 2;

[0083] Among them, L f L is the distance from the fault point to a certain end. line v is the total length of the line. wave Let Δt be the propagation speed of the traveling wave in the circuit. arrival This indicates the time difference between the arrival of the high-frequency traveling wave signal generated by the fault at the measuring devices at both ends of the line.

[0084] Alternatively, an AI-based fault location method can be used. For example, graph neural networks can be used to analyze the network's measurement data to directly output the faulty line and its approximate location, or the slight difference in the arrival time of the current waveforms on both sides of the differential protection unit can be used for auxiliary judgment. The output of this step is precise fault location information.

[0085] Step 4.3: Output the adjusted protection settings and fault location results;

[0086] Adjust the specific protection settings (such as the updated I) op_thresh and k slope The fault location results (such as the distance to the fault point) are sent to the corresponding protection device and reported to the dispatch system or maintenance personnel.

[0087] The strategy execution and optimization module is used to execute collaborative scheduling strategies online and to verify and continuously optimize the strategies.

[0088] Specifically, the following steps are included:

[0089] Step 5.1: Execute the strategy online;

[0090] The DRL collaborative scheduling strategy is deployed into the distribution network dispatching system. The system adjusts the system based on real-time status s. t Action a is obtained through DRL model reasoning. t This refers to specific protection setting adjustment commands and active management control commands, which are then sent to the corresponding protection devices and execution units such as DG controllers and energy storage controllers.

[0091] Step 5.2, Strategy Verification;

[0092] The effectiveness and security of online scheduling strategies are verified in real-time or near real-time using a digital twin simulation platform. This digital twin simulation platform is a highly realistic virtual environment of a power distribution network, and its core components include:

[0093] High-precision power grid model units can accurately simulate the physical characteristics of distribution networks, such as topology, line parameters, transformer models, and load models.

[0094] The distributed power generation model unit is capable of simulating the dynamic behavior and control logic of various distributed generation (DG) types such as IIDG and SGDG.

[0095] The fault simulation unit is capable of simulating short-circuit faults and other disturbance events of various types and locations;

[0096] The data interaction interface unit is used to receive real-time data from the actual system or control commands from the DRL agent and output simulation results.

[0097] This platform allows for comprehensive testing and evaluation of the scheduling strategies generated by the DRL agent without affecting the actual operation of the power grid. For example, it can simulate scenarios where a certain IIDG fails and goes out of service, and whether the DRL strategy can promptly adjust the differential protection settings of adjacent lines and instruct other available DGs or energy storage systems to provide power support to avoid cascading failures or large-scale power outages.

[0098] The performance of the scheduling strategy can be evaluated by comparing system responses (such as fault clearing time, DG disconnection rate, and network loss) with and without the strategy. Alternatively, practical application verification can be conducted in a small-scale pilot area where conditions permit.

[0099] Step 5.3, iterative optimization of the DRL model;

[0100] The system possesses continuous learning capabilities. It collects actual operational data, protection action information, DG response data, and verification results during the online execution of scheduling strategies.

[0101] When the accumulated data reaches a certain amount, or when a major power grid event occurs (such as new distributed generation (DG) grid connection, network topology changes, or a significant decline in protection strategy evaluation indicators), the DRL model is retrained and its parameters are optimized to adapt to the changing environment and improve strategy performance. The optimized model is then redeployed, forming a closed-loop continuous learning and optimization mechanism.

[0102] Application example of this implementation method:

[0103] Application scenarios:

[0104] Consider a 10kV distribution network feeder that includes multiple types of distributed generation, such as a photovoltaic power station connected in the middle of the feeder and a small gas turbine (SGDG, Synchronous Generator-based Distributed Generation) connected at the end.

[0105] The feeder uses differential line protection as its main protection. During the peak load period in the summer evenings, the photovoltaic output decreases, while the gas turbine operates stably as planned. At this time, if a metallic three-phase short-circuit fault occurs in the feeder...

[0106] Implementation process example

[0107] Real-time acquisition of multi-source heterogeneous data and panoramic status perception: The system acquires in real time voltage and current waveform data uploaded by each PMU, switch status and load data fed back by the SCADA system, and real-time output P from the photovoltaic power station. PV DC side voltage U dc The rotational speed ω of the gas turbine SGDG Excitation current I f Data such as these are used to create a snapshot of the power grid's operation at the current moment.

[0108] Online identification of hybrid DG dynamic fault characteristics:

[0109] For photovoltaic power plants: By analyzing the voltage and current transient waveforms collected by the PMU before and after the fault, and using online parameter identification algorithms (such as the least squares-based recursive algorithm), it can be quickly estimated that the fault is equivalent to a controlled current source, with its maximum fault current contribution being about 1.2 times its rated current, and the duration not exceeding 20ms.

[0110] For gas turbines: also based on fault transient data, it is identified that they exhibit subtransient characteristics in the initial stage of a fault, and their equivalent subtransient reactance X d "Approximately 0.15 pu, capable of providing a short-circuit current of about 3 to 5 times the rated current. These identified parameters I..." IIDG_fault_max T IIDG_inject X d_SGDG "It is passed to subsequent modules. Among them, I" IIDG_fault_max This indicates the maximum fault current amplitude that an inverter-interface type distributed power source can provide under fault conditions, usually expressed as a multiple of the rated current; T IIDG_inject Indicates the duration for which the IIDG can continuously inject fault current during a fault; X d_SGDG "" represents the subtransient reactance value of a synchronous generator-type distributed power source, which is a key parameter that determines the magnitude of the short-circuit current in the initial stage of an SGDG fault.

[0111] Generation of integrated protection and control collaborative scheduling strategies based on deep reinforcement learning:

[0112] The DRL agent receives the current grid status (including heavy line load and low photovoltaic output) and identifies the current-limiting characteristics of the IIDG and the strong fault current support capability of the SGDG. Based on the learned strategy, the agent outputs the following cooperative instructions:

[0113] Protection setting adjustment command: For the line differential protection of this feeder, considering that the current limiting characteristic of IIDG may lead to a decrease in differential current, the DRL intelligent agent command appropriately reduces the operating current threshold I of the differential protection. op_thresh The current has been adjusted from 0.5A (secondary side) to 0.3A to improve protection sensitivity and ensure reliable fault detection.

[0114] Active distribution network management measures: Since the photovoltaic output is already low and the gas turbine has a certain ramping capability, the DRL agent instructs the gas turbine to temporarily and appropriately increase the active power output ΔP within the allowable range. SGDG This is intended to help maintain system frequency and voltage stability after the fault is cleared, thereby reducing the impact on the main grid.

[0115] Adaptive online setting and precise fault location for specific protection systems:

[0116] The differential protection device receives a new operating current threshold Iop_thresh =0.3A and updated online. When a fault occurs, because the setting has been adapted to the weak feedback characteristics of IIDG, the differential protection can act quickly and accurately, issuing a trip command to disconnect the faulty line.

[0117] Meanwhile, traveling wave acquisition units deployed at both ends of the line detected the fault traveling wave and calculated the arrival time difference Δt of the traveling wave. arrival Combined with line parameters (line length L) line =10km, wave speed v wave =2.8×10 5 km / s), if Δt is measured arrival =10μs, then the distance from the fault point to one end is calculated to be L. f =(10-2.8×10) 5 ×10×10 -6 ) / 2=(10-2.8) / 2=3.6km.

[0118] Online execution, verification, and continuous iterative optimization of scheduling strategies: Reporting of protection actions and fault location results. During subsequent operation, the system continuously collects data. If it finds that protection actions are unsatisfactory or DRL strategy outputs are deviated in similar scenarios, it will trigger model retraining.

[0119] Technical effectiveness verification:

[0120] Through the above embodiments, the technical solution can achieve the following main technical effects:

[0121] Improving the accuracy of protection actions: Under traditional fixed settings, differential protection may fail to operate due to insufficient differential current caused by the current-limiting characteristics of IIDG. This solution identifies IIDG fault characteristics online and adaptively adjusts the differential protection operating threshold (e.g., from 0.5A to 0.3A), ensuring reliable protection operation under weak IIDG feeder faults and avoiding the expansion of the fault range and equipment damage caused by failure to operate.

[0122] Compared with traditional fixed values, the accuracy of protection actions is significantly improved in fault scenarios involving IIDG.

[0123] Improving fault location accuracy: Traditional fault indicators or impedance-based ranging methods may experience a decrease in accuracy when DG (Distributed Gauge) is integrated (especially IIDG). This solution combines high-frequency traveling wave data with precise Δt... arrival Analysis shows that it can achieve meter-level fault location, which is far superior to the traditional hundred-meter or even kilometer-level positioning accuracy, thus greatly shortening the fault troubleshooting time and the power outage area.

[0124] Tables 1 and 2 present a comparison of simulated data, demonstrating the effectiveness of this technical solution in improving protection sensitivity and fault location accuracy:

[0125] Table 1: Comparison of Protective Action Sensitivity;

[0126]

[0127] Table 2: Comparison of fault location accuracy;

[0128] Fault location methods Actual location of the fault (km) Location results (km) Absolute error (m) Traditional impedance ranging method 3.600 3.950 350 This scheme uses the traveling wave positioning method. 3.600 3.605 5

[0129] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. An adaptive relay protection setting online scheduling system with distributed power supply, characterized in that, include: The data acquisition module is used to collect multi-source heterogeneous data from the power distribution network in real time and to build a panoramic view of the power distribution network operation. The fault characteristic identification module is used to identify the dynamic fault characteristic parameters of hybrid distributed generation in the distribution network online. This module utilizes high-frequency real-time data and employs at least one parameter identification algorithm selected from least squares method, Kalman filtering, or particle swarm optimization to estimate the equivalent Thevenin impedance of at least one of the inverter-interface type distributed generation and the synchronous generator type distributed generation online. The equivalent Thevenin impedance is... ,in This represents the equivalent internal impedance of a distributed power source. Equivalent resistance For equivalent reactance, The weighting coefficients represent the equivalent reactance; the fault characteristic identification module also constructs at least one machine learning model among recurrent neural networks, long short-term memory networks, or gated recurrent units, inputting historical fault data and real-time operating data to predict the dynamic response characteristics of different types of distributed power sources under different fault types and fault locations, including the amplitude of the fault current. Phase angle and the duration of fault current injection ; The collaborative scheduling strategy generation module, based on a deep reinforcement learning algorithm, generates an integrated protection and control collaborative scheduling strategy according to a panoramic operation view and dynamic fault characteristic parameters. The collaborative scheduling strategy includes protection setting adjustment instructions and control instructions for proactive distribution network management measures. This module constructs the distribution network operation and protection setting scheduling problem as a Markov decision process. The state input includes at least one of real-time grid measurements, distributed generation status, and current protection settings. The action output includes at least one of protection setting adjustment instructions and control instructions for proactive distribution network management measures. The reward function comprehensively considers at least one of the following: distribution network operation economy, distributed generation absorption capacity, system safety margin, and penalties for improper protection behavior. The reward function is... ,in Let be the reward function at time t. For economic allocation benefits, For network loss, This refers to the capacity for distributed power generation. To protect the penalties for misconduct, , , , These are the weight coefficients for economic dispatch benefits, network losses, distributed power consumption, and penalties for improper protection behavior, respectively. The cooperative dispatch strategy generation module uses a deep reinforcement learning algorithm to train the agent to generate a cooperative dispatch strategy. The deep reinforcement learning algorithm includes an Actor network and a Critic network. Accept the current state As input, output action ,in This represents the system state at the current time t. For the parameter set of the Actor network, The policy function; the Critic network Enter the current status and corresponding actions Output the value estimate of the state-action pair, where This represents the set of parameters for the Critic network. The action value function; The adaptive setting and fault location module is used to adaptively set the online settings of specific protection based on dynamic fault characteristic parameters and collaborative scheduling strategies. It also combines high-frequency sampling data analysis technology to achieve precise fault location. The module adjusts the differential protection's operating current threshold online based on the dynamic fault current contribution characteristics of the distributed power sources connected at both ends of the line under different operating modes and the setting adjustment direction given by the collaborative scheduling strategy. Sum of ratio braking coefficient When it is identified that the inverter interface-type distributed power source exhibits current-limiting characteristics in the early stages of a fault, reduce... Or adjust The inflection point; the adaptive setpoint tuning and fault location module adopts an algorithm based on traveling wave ranging, which analyzes the time difference between the arrival of the high-frequency traveling wave signal generated at the moment of the fault at the measuring devices at both ends of the line. The distance to the fault point is calculated by combining the line parameters. The calculation formula is as follows: ,in The distance from the fault point to a certain end. This is the total length of the line. Let $\frac{ ... The time difference between the arrival of the high-frequency traveling wave signal generated by the fault at the measuring devices at both ends of the line; The strategy execution and optimization module is used to execute collaborative scheduling strategies online and to verify and continuously iterate the optimization of these strategies. This module utilizes a digital twin simulation platform to verify the effectiveness and security of the collaborative scheduling strategies. The digital twin simulation platform includes a high-precision power grid model unit, a distributed power source model unit, a fault simulation unit, and a data interaction interface unit. The high-precision power grid model unit accurately simulates the topology, line parameters, transformer models, and load models of the distribution network. The distributed power source model unit simulates the dynamic behavior and control logic of inverter-interface type distributed power sources and synchronous generator type distributed power sources. The fault simulation unit simulates various types and locations of short-circuit faults and other disturbance events. The data interaction interface unit receives real-time data from the actual system or control commands from the deep reinforcement learning agent and outputs simulation results. The strategy execution and optimization module utilizes accumulated operational data and scheduling experience. When the accumulated data reaches a certain amount, or when at least one major power grid event occurs—such as new distributed power source grid connection, network topology change, or a significant decline in protection strategy evaluation indicators—it triggers retraining and iterative optimization of the deep reinforcement learning algorithm model.

2. The adaptive relay protection setting online scheduling system with distributed power supply according to claim 1, characterized in that, The data acquisition module collects multi-source heterogeneous data, including: At least one of the following: synchronous phasor measurement unit data, SCADA system data, operating status parameters of various distributed power sources, protection device information, and meteorological data.

3. The system for on-line scheduling of relay settings for adaptive protection with distributed generation of claim 1, wherein, The deep reinforcement learning algorithm is an Actor-Critic type algorithm.

4. The adaptive relay protection setting online scheduling system with distributed power supply according to claim 1, characterized in that, Control commands for active distribution network management measures include at least one of the following: active or reactive power regulation commands for distributed generation and energy storage charging and discharging commands.

5. A computer readable storage medium, characterized in that, It is used to store computer-readable instructions, which, when read by a computer, enable the operation of an adaptive relay protection setting online scheduling system with distributed power supply as described in any one of claims 1 to 4.

Images