Equivalent parameter identification method for minimizing generalization error in multi-disturbance scenarios

Abstract

The application discloses an equivalent parameter identification method for minimizing generalization error of multi-disturbance scenes, comprising the following steps: acquiring topology and operation data of an active power distribution network, and constructing a time-domain simulation model of the power distribution network; constructing an electromechanical equivalent model according to the time-domain simulation model and initializing the electromechanical equivalent model; screening a to-be-identified parameter vector through transient segmented trajectory sensitivity analysis; generating a fault scene set according to the time-domain simulation model; for each fault scene in the fault scene set, respectively adopting a two-stage hybrid optimization strategy to assign and optimize the to-be-identified parameter vector, and obtaining a candidate parameter vector set; calculating a multi-scene generalization error of each candidate parameter vector; and taking the candidate parameter vector with the minimum error as the final parameter vector of the electromechanical equivalent model. The application realizes efficient, high-precision and strong generalization capability of external disturbance for photovoltaic power distribution network equivalent parameter identification.

Description

Technical Field

[0001] This invention relates to an equivalent parameter identification method for minimizing generalization error in multi-perturbation scenarios, and belongs to the field of parameter identification technology. Background Technology

[0002] The high proportion of distributed photovoltaic (PV) grid connections means that the transient characteristics of the distribution network are jointly dominated by PV control strategies and induction motor loads. Due to the large number of nodes in the distribution network and the massive computational demands of detailed modeling simulations, it is typically necessary to aggregate them to a 110kV point of common coupling (PCC) to construct a simplified electromechanical transient model to meet the simulation requirements of large power grids. However, existing equivalent modeling and parameter identification techniques for distribution networks containing distributed PV still have the following limitations:

[0003] 1) The parameters to be identified are highly dimensional and physically coupled. Existing methods do not divide the electromechanical transient process into stages and lack decoupling analysis of the sensitivity of active and reactive power trajectories. This makes it impossible to clarify the dominant role of each parameter in different transient stages. All parameters participate in optimization indiscriminately, increasing computing power and affecting identification accuracy.

[0004] 2) Traditional single heuristic algorithms are prone to premature convergence when solving high-dimensional nonlinear spaces, getting stuck in local optima, and it is difficult to meet the requirements of global exploration and high-precision local convergence.

[0005] 3) Existing studies mostly rely on the response under a single power grid disturbance scenario for parameter identification. Due to the complex coupling between multidimensional parameters, "different parameters with the same effect" and overfitting are prone to occur, making it difficult for equivalent models to adapt to unknown disturbances and changes in the operation mode of the external power grid. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an equivalent parameter identification method for minimizing generalization error in multi-disturbance scenarios. By reducing the identification dimension through transient segmented trajectory sensitivity analysis, improving the two-stage optimization of particle swarm optimization and simplex method, and verifying parameters based on multi-scenario generalization error, this invention achieves efficient, high-precision, and highly generalizable equivalent parameter identification for photovoltaic distribution networks that is resistant to external disturbances.

[0007] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0008] This invention discloses an equivalent parameter identification method for minimizing generalization error in multi-perturbation scenarios, including:

[0009] Obtain the topology and operation data of an active power distribution network containing high-penetration distributed photovoltaic power; construct a time-domain simulation model of the power distribution network based on the topology and operation data;

[0010] Based on the time-domain simulation model, an electromechanical equivalent model is constructed; the electromechanical equivalent model is initialized based on the network loss consistency principle and the capacity weighting principle to obtain the initialized electromechanical equivalent model;

[0011] Based on the initialized electromechanical equivalent model, the parameter vector to be identified is obtained through transient segmented trajectory sensitivity analysis.

[0012] Based on the time-domain simulation model, a set of fault scenarios is generated. For each fault scenario in the set of fault scenarios, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is used to assign and optimize the parameter vector to be identified, thereby obtaining the candidate parameter vector under each fault scenario. The candidate parameter vector set is composed of all the candidate parameter vectors.

[0013] Substitute the candidate parameter vector set into the fault scenario set to perform electromechanical transient simulation, and calculate the multi-scenario generalization error of each candidate parameter vector set.

[0014] The candidate parameter vector with the smallest generalization error in the multi-scenario model is selected as the final parameter vector of the electromechanical equivalent model, thereby realizing the identification of equivalent parameters for minimizing the generalization error in multi-disturbance scenarios.

[0015] Furthermore, the construction of the time-domain simulation model of the power distribution network includes:

[0016] Based on the topology, determine the access node locations of static loads, dynamic loads, compensation capacitors, and photovoltaics, as well as the electrical connection relationships between distribution network lines and transformers;

[0017] Based on the aforementioned operational data, mathematical models for photovoltaic, static load, and dynamic load are established respectively.

[0018] Based on the mathematical models of photovoltaic, static load and dynamic load, interconnection is performed according to the location and electrical connection relationship of the access node, and the reactive power compensation capacity parameters of the compensation capacitor and the impedance parameters of the distribution network lines and transformers are connected at the corresponding access node to construct a time-domain simulation model of the distribution network.

[0019] Furthermore, the mathematical model of the photovoltaic system includes a mathematical model for type A photovoltaic system and a mathematical model for type B distributed photovoltaic system. The mathematical model for type A photovoltaic system adopts fault ride-through control logic that includes calculations of active current commands and reactive current commands during low voltage ride-through. The mathematical model for type B distributed photovoltaic system adopts a blocking control logic that includes power attenuation logic under blocking control during faults and power linear recovery logic after fault clearance.

[0020] The mathematical model of the static load adopts a polynomial static load model that includes constant impedance components, constant current components, and constant power components.

[0021] The mathematical model of the dynamic load uses a third-order electromechanical transient model to characterize the dynamic response of the induction motor group.

[0022] Furthermore, the construction of the electromechanical equivalent model includes:

[0023] In the time-domain simulation model, the static load, dynamic load, compensation capacitor, and photovoltaic are respectively represented as equivalent components, and all equivalent components are connected to the same virtual bus. The virtual bus is connected to the point of common connection via an equivalent impedance. The equivalent components include equivalent static load, equivalent dynamic load, equivalent capacitor, and equivalent photovoltaic.

[0024] Furthermore, obtaining the initialized electromechanical equivalent model includes:

[0025] Based on the reactive power of each compensation capacitor in the time-domain simulation model, the capacitance value of the equivalent capacitor is calculated.

[0026] Based on the rated power and voltage characteristic coefficients of each static load in the time-domain simulation model, the equivalent rated power and voltage characteristic coefficients of the equivalent static load are calculated according to the capacity weighting principle.

[0027] Based on the electrical parameters, mechanical parameters, and rated capacity of each dynamic load in the time-domain simulation model, the equivalent electrical parameters and equivalent mechanical parameters of the equivalent dynamic load are calculated according to the capacity weighting principle.

[0028] Based on the rated power of the photovoltaic and the active power loss of the network in the time-domain simulation model, the equivalent rated power of the equivalent photovoltaic is calculated.

[0029] Based on the power conservation relationship between the virtual bus and the common connection point, the resistance and reactance values ​​of the equivalent impedance are calculated according to the principle of consistent network loss, thus completing the initialization of the electromechanical equivalent model.

[0030] Furthermore, the filtering process yields a vector of parameters to be identified, including:

[0031] Determine multiple dynamic parameters of the initialized electromechanical equivalent model;

[0032] The transient response process of the initialized electromechanical equivalent model is divided into a fault period and a recovery period. By calculating the active power sensitivity, reactive power sensitivity, active power sensitivity, and reactive power sensitivity of each dynamic parameter during the fault period, the global comprehensive sensitivity of each dynamic parameter is obtained.

[0033] A preset number of dynamic parameters ranked first in global comprehensive sensitivity are selected to form the parameter vector to be identified, while the remaining dynamic parameters are taken as fixed values.

[0034] Furthermore, the generation of the fault scenario set includes:

[0035] At the common connection point of the time-domain simulation model, a voltage drop set containing different target drop depths is constructed according to a preset voltage step size;

[0036] For each target dip depth in the voltage dip set, the short-circuit fault conditions in the external transmission network are adjusted until the dip voltage at the point of common coupling during the fault reaches the target dip depth, and the corresponding short-circuit fault conditions are taken as the fault scenarios corresponding to the target dip depth.

[0037] By iterating through all target drop depths, a set of failure scenarios consisting of multiple failure scenarios is obtained.

[0038] Furthermore, the two-stage hybrid optimization strategy includes:

[0039] For each fault scenario in the fault scenario set, an adaptive particle swarm optimization algorithm is used for global search to obtain a preliminary solution for the parameter vector to be identified.

[0040] Using the preliminary solution of the parameter vector to be identified as the initial point, the Nelder-Mead simplex method is used for local search to obtain the candidate parameter vector.

[0041] The adaptive particle swarm optimization algorithm and the Nelder-Mead simplex method both aim to minimize the error function to optimize the vector of parameters to be identified. The error function is used to characterize the difference in power response between the electromechanical equivalent model and the time-domain simulation model at the common connection point.

[0042] Furthermore, the expression for the error function is as follows:

[0043]

[0044] In the formula, Represents the parameter vector to be identified The error function, , Indicates the number of parameters to be identified;

[0045] Indicates the weighting coefficient during the fault period;

[0046] The weighting coefficient represents the active power error during a fault.

[0047] The weighting coefficient represents the reactive power error during a fault.

[0048] This indicates the active power error during the fault period;

[0049] This indicates the reactive power error during the fault period;

[0050] Indicates the weighting coefficient during the recovery period;

[0051] The weighting coefficient represents the active power error during the recovery period;

[0052] The weighting coefficient represents the reactive power error during the recovery period;

[0053] This indicates the active power error during the recovery period;

[0054] This indicates the reactive power error during the recovery period.

[0055] Furthermore, the calculation of the multi-scenario generalization error for each set of candidate parameter vectors includes:

[0056] For each set of candidate parameter vectors in the candidate parameter vector set, substitute them into each fault scenario in the fault scenario set to perform electromechanical transient simulation, and obtain the active power simulation value and reactive power simulation value of the electromechanical equivalent model at the common connection point under each fault scenario;

[0057] Obtain the active power response value and reactive power response value of the time-domain simulation model under the fault scenario;

[0058] Based on the simulated active power value, simulated reactive power value, active power response value, and reactive power response value, calculate the root mean square error of each candidate parameter vector under each fault scenario.

[0059] The root mean square error of each candidate parameter vector under each fault scenario is weighted and summed to obtain the multi-scenario generalization error of each candidate parameter vector.

[0060] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0061] The equivalent parameter identification method of the present invention for minimizing the generalization error of multi-disturbance scenarios firstly divides the electromechanical transient response into fault period and recovery period through transient segmented trajectory sensitivity analysis, decouples the calculation of the stage sensitivity of active power and reactive power to each dynamic parameter, and selects the parameter vector to be identified accordingly, which effectively reduces the dimensionality and calculation time of parameter identification and improves the calculation efficiency of electromechanical transient simulation.

[0062] Secondly, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is adopted to assign values ​​to the vector of parameters to be identified for optimization. The improved particle swarm optimization algorithm achieves a global search to obtain a preliminary solution through adaptive inertia weights and time-varying learning factors. The Nelder-Mead simplex method uses this preliminary solution as the initial point to perform a local fine search, which takes into account both global exploration capability and local convergence accuracy, avoids the problem of single algorithms easily getting trapped in local optima, and improves the accuracy of isoparameter identification.

[0063] Finally, by substituting the candidate parameter vector set into the fault scenario set for electromechanical transient simulation, the final parameter vector of the electromechanical equivalent model is selected based on the criterion of minimizing the generalization error of multiple scenarios. This effectively overcomes the problems of heterogeneous parameters with the same effect and overfitting caused by single-scenario identification, enabling the electromechanical equivalent model to adapt to external disturbances of different positions and intensities, and significantly improving the model's generalization ability. Attached Figure Description

[0064] Figure 1 This is a flowchart of the equivalent parameter identification method for minimizing generalization error in multi-perturbation scenarios provided in an embodiment of the present invention;

[0065] Figure 2 This is a schematic diagram of the structure of the photovoltaic power generation system model provided in the embodiment of the present invention;

[0066] Figure 3 This is a block diagram of VSC grid connection control provided in an embodiment of the present invention;

[0067] Figure 4 This is a high / low voltage ride-through requirement curve for distributed photovoltaic systems provided in an embodiment of the present invention;

[0068] Figure 5 This is a flowchart of the converter protection strategy during fault ride-through provided in an embodiment of the present invention;

[0069] Figure 6 This is a schematic diagram of typical transient characteristics of Class A photovoltaic provided in an embodiment of the present invention;

[0070] Figure 7 This is a schematic diagram of typical transient characteristics of type B photovoltaic provided in an embodiment of the present invention;

[0071] Figure 8This is an equivalent circuit diagram of a third-order induction motor model provided in an embodiment of the present invention;

[0072] Figure 9 This is a structural diagram of the electromechanical equivalent model of the power distribution network provided in the embodiments of the present invention;

[0073] Figure 10 This is a schematic diagram of a stage division scheme based on the transient response of an active power distribution network provided in an embodiment of the present invention. Detailed Implementation

[0074] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0075] This embodiment provides an equivalent parameter identification method for minimizing generalization error in multi-perturbation scenarios, such as... Figure 1 As shown, it includes:

[0076] Obtain the topology and operation data of an active power distribution network with high-penetration distributed photovoltaic power; construct a time-domain simulation model of the power distribution network based on the topology and operation data;

[0077] Based on the time-domain simulation model, an electromechanical equivalent model is constructed; the electromechanical equivalent model is initialized based on the principle of consistent network loss and the principle of capacity weighting, resulting in the initialized electromechanical equivalent model;

[0078] Based on the initialized electromechanical equivalent model, the parameter vector to be identified is obtained through transient segmented trajectory sensitivity analysis.

[0079] Based on the time-domain simulation model, a set of fault scenarios is generated. For each fault scenario in the set, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is adopted to assign and optimize the parameter vector to be identified, thereby obtaining the candidate parameter vector under each fault scenario. The candidate parameter vector set is composed of all the candidate parameter vectors.

[0080] Substitute the candidate parameter vector set into the fault scenario set to perform electromechanical transient simulation, and calculate the multi-scenario generalization error of each candidate parameter vector set.

[0081] The candidate parameter vector with the smallest generalization error across multiple scenarios is selected as the final parameter vector of the electromechanical equivalent model, thereby achieving equivalent parameter identification for minimizing the generalization error in multi-disturbance scenarios.

[0082] The technical concept of this invention is as follows:

[0083] First, by performing transient segmented trajectory sensitivity analysis, the electromechanical transient response is divided into the fault period and the recovery period. The stage sensitivity of active power and reactive power to each dynamic parameter is decoupled and calculated. Based on this, the parameter vector to be identified is selected, which effectively reduces the dimensionality of parameter identification and the calculation time, and improves the calculation efficiency of electromechanical transient simulation.

[0084] Secondly, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is adopted to assign values ​​to the vector of parameters to be identified for optimization. The improved particle swarm optimization algorithm achieves a global search to obtain a preliminary solution through adaptive inertia weights and time-varying learning factors. The Nelder-Mead simplex method uses this preliminary solution as the initial point to perform a local fine search, which takes into account both global exploration capability and local convergence accuracy, avoids the problem of single algorithms easily getting trapped in local optima, and improves the accuracy of isoparameter identification.

[0085] Finally, by substituting the candidate parameter vector set into the fault scenario set for electromechanical transient simulation, the final parameter vector of the electromechanical equivalent model is selected based on the criterion of minimizing the generalization error of multiple scenarios. This effectively overcomes the problems of heterogeneous parameters with the same effect and overfitting caused by single-scenario identification, enabling the electromechanical equivalent model to adapt to external disturbances of different positions and intensities, and significantly improving the model's generalization ability.

[0086] The specific steps are as follows:

[0087] Step 1: Obtain the topology and operation data of the active power distribution network containing high-penetration distributed photovoltaic power; based on the topology and operation data, construct a time-domain simulation model of the power distribution network.

[0088] Constructing a time-domain simulation model of the power distribution network includes:

[0089] Based on the topology, determine the access node locations for static loads, dynamic loads, compensation capacitors, and photovoltaics, as well as the electrical connection relationships between distribution network lines and transformers;

[0090] Based on the operational data, mathematical models for photovoltaic, static load, and dynamic load were established respectively.

[0091] Based on the mathematical models of photovoltaics, static loads, and dynamic loads, interconnections are made according to the location of access nodes and electrical connections. The reactive power compensation capacity parameters of the compensation capacitors and the impedance parameters of the distribution network lines and transformers are connected at the corresponding access nodes to construct a time-domain simulation model of the distribution network.

[0092] The mathematical models for photovoltaics include the mathematical model for Class A photovoltaics and the mathematical model for Class B distributed photovoltaics. The mathematical model for Class A photovoltaics uses fault ride-through control logic that includes calculations of active current commands and reactive current commands during low voltage ride-through. The mathematical model for Class B distributed photovoltaics uses a shielding control logic that includes power attenuation logic under shielding control during faults and power linear recovery logic after fault clearance.

[0093] The mathematical model of static load adopts a polynomial static load model that includes constant impedance components, constant current components, and constant power components.

[0094] The mathematical model of dynamic load uses a third-order electromechanical transient model to characterize the dynamic response of the induction motor group.

[0095] The specific construction of the mathematical model is as follows:

[0096] 1.1 Mathematical Model of Photovoltaics

[0097] Single-stage photovoltaic power generation system topology as follows Figure 2 As shown in the figure, , These are the DC voltage and current output by the photovoltaic cell, respectively. It is a DC capacitor; The input current to the converter is the DC current; the converter VSC (Voltage Source Converter) control system is the core module of the photovoltaic power generation system, mainly used to convert the DC current generated by the photovoltaic array into AC current with the same frequency and phase as the AC grid voltage, and then output it after filtering and connecting it to the grid. , These are the inductance and capacitance of the filter, respectively; , These are the AC side current and voltage of the inverter, respectively. This refers to the grid-side voltage. This is the equivalent reactance of the photovoltaic output line.

[0098] VSC is a three-phase bridge converter device composed of fully controlled power electronic devices. For grid connection control of VSC under normal operating conditions, see [link to relevant documentation]. Figure 3 The grid-connected control system adopts a dual-loop controller structure combining an outer loop and an inner loop. The outer loop control realizes the active and reactive power control of the VSC and determines the current reference value, while the inner loop control realizes the rapid tracking of the output current to the reference current.

[0099] The outer loop active power control uses constant DC side voltage control, and the voltage outer loop uses the maximum power point voltage obtained by photovoltaic maximum power point tracking (MPPT) technology. PI control is used as a reference value to stabilize the DC-side voltage of the VSC. The reactive power in the outer loop is controlled by constant power factor.

[0100] The inner current loop adopts a dual closed-loop control strategy in a synchronous rotating coordinate system, utilizing coordinate transformation to convert the three-phase AC quantities at the VSC grid connection point ( , , , , , Convert to synchronous coordinate system , Direct flow ( , , , VSC active power output is from Shaft current control, its reference value The output of the outer voltage loop; the reactive power output of VSC is... Shaft current control. The current reference command is determined using power factor mode. VSC typically operates in unity power factor mode, i.e. The value is set to 0. The current inner loop output signal is transformed by coordinate transformation and space vector pulse width modulation (SVPWM) to obtain the VSC switch drive signal. … .

[0101] To achieve orientation of the grid voltage vector during coordinate transformation, a phase-locked loop (PLL) technique based on the grid voltage fundamental frequency is typically incorporated to obtain the position angle of the voltage vector. Then, perform coordinate transformation.

[0102] According to the "Technical Regulations for Grid Connection of Photovoltaic Power Stations," photovoltaic power stations connected to the grid at voltage levels of 35kV and above must possess High Voltage Ride Through (HVRT) and Low Voltage Ride Through (LVRT) capabilities. Specific requirements are detailed in [link to relevant regulations]. Figure 4 It also proposes the dynamic reactive current requirements injected into the grid by the photovoltaic power station during fault ride-through, as shown in equation [equation missing]. ,Mode :

[0103]

[0104]

[0105] In the formula, This represents the reference value of reactive current in a synchronous rotating coordinate system.

[0106] This represents the reference value of active current in a synchronous rotating coordinate system.

[0107] This represents the per-unit value of the voltage at the photovoltaic grid connection point.

[0108] The overall process of the inverter protection strategy during fault ride-through is as follows: Figure 5 Based on the grid connection point voltage During normal operation ( reactive current command Voltage drop ( ) or temporary promotion ( When the total current command exceeds the limit (>1.05 times), the reactive current command is injected or absorbed at a factor of 1.5. During low-voltage ride-through, if the total current command exceeds the limit (>1.05 times), the active current command is limited according to the "reactive priority" strategy. This is to achieve overcurrent protection. The final current command is sent to the inner loop control for execution.

[0109] Based on whether they have voltage ride-through capability, photovoltaic power generation system models are divided into Class A photovoltaic models and Class B photovoltaic models.

[0110] like Figure 6 As shown, under normal conditions, the transient characteristics of Class A photovoltaic (PV) systems are similar to constant power control, and sometimes constant DC voltage control is also used. During ride-through conditions, the characteristics of PV systems from different manufacturers and models generally differ, but they can generally be modeled as current or power sources affected by the generator terminal voltage. Most PV systems operate in a specified current control mode, with the current command as follows:

[0111]

[0112] In the formula, This indicates the active current command during the low-voltage ride-through of new energy sources. This indicates the reactive current command during the low-voltage ride-through period of new energy sources.

[0113] Indicates the rated capacity of new energy sources; This represents the active current-terminal voltage droop factor. The per-unit value representing the voltage at the photovoltaic grid connection point;

[0114] This represents the active current bias coefficient considering the crossover state. This represents the active current before the new energy source enters the low-voltage ride-through phase. This represents the active current bias coefficient during the crossing period;

[0115] This represents the reactive current-terminal voltage droop factor. This represents the reactive current bias coefficient considering the state before crossing; This represents the reactive current bias coefficient during the ride-through period; represents the reactive current of the new energy source before it enters the low voltage ride-through stage; represents the low voltage ride-through judgment threshold, which is set to 0.9 pu in this embodiment.

[0116] The specified current control mode directly controls the current output. During voltage dips, reactive current is preferentially injected to boost the grid connection voltage. In the specified power control mode, to maintain power output during voltage drops, the current needs to increase, potentially triggering overcurrent protection. When performing aggregated equivalent modeling of distributed photovoltaic systems in the distribution network, this model only considers the specified current control mode under low-voltage ride-through conditions.

[0117] like Figure 7 As shown, Class B distributed renewable energy is a full-power VSC grid-connected renewable energy source without low-voltage ride-through capability. Compared to Class A photovoltaic, Class B photovoltaic is not designed and manufactured with the operational requirements of centralized renewable energy sources in mind, and lacks low-voltage ride-through control measures. However, it is equipped with a "wave-blocking (soft grid disconnection)" function for protection equipment. When the voltage is low, in order to protect the converter equipment, the renewable energy source will stop power output, and after a period of time, it will slowly recover to the state before the wave-blocking. The power recovery process is generally a linear recovery with a fixed slope.

[0118] At low voltage, surge protection control is used instead of low voltage ride-through control. Assume that the Class B converter starts surge protection at time t=0, and the power during the surge protection period is calculated according to the formula... and formula calculate:

[0119]

[0120]

[0121] In the formula, This indicates the active power during the wave blocking period; This indicates the reactive power during the wave blocking period;

[0122] This indicates the active power before entering the wave blocking phase;

[0123] Indicates the duration of the wave-blocking state;

[0124] Indicates the recovery rate of active power;

[0125] 1.2 Mathematical Model of Static Load

[0126] Based on the characteristics of load power changing with voltage, static loads are divided into three categories: constant impedance, constant current, and constant power loads.

[0127]

[0128]

[0129] In the formula, This represents the actual active power of the load node; This represents the actual reactive power of the load node;

[0130] This represents the active power capacity of the static load under rated operating conditions. This represents the reactive power capacity of the static load under rated operating conditions;

[0131] This indicates the actual operating voltage amplitude at the load connection node; This indicates the rated voltage amplitude at the load connection node;

[0132] This indicates the proportional weight of constant impedance in the active load;

[0133] This indicates the proportional weight of constant current in the active load;

[0134] This indicates the proportion and weight of constant power in the active load. ;

[0135] This indicates the proportional weight of constant impedance in reactive loads;

[0136] This indicates the proportional weight of constant current in reactive loads;

[0137] This indicates the proportion and weight of constant power in reactive load. ;

[0138] This represents the difference between the system's actual operating frequency and its rated frequency.

[0139] This represents the adjustment coefficient of the system frequency fluctuation on the active power load.

[0140] This represents the adjustment coefficient of reactive load power to the system frequency fluctuation.

[0141] 1.3 Mathematical Model of Dynamic Load

[0142] In the dynamic load equivalent modeling stage, a third-order electromechanical transient model is used to represent the dynamic response of the induction motor group. The equivalent circuit is as follows: Figure 8 As shown,

[0143] In the picture, Indicates the resistance of the stator; Indicates the resistance of the rotor; Represents the reactance of the stator; Indicates the reactance of the rotor; Indicates the magnetizing reactance; This indicates the rotor slip rate.

[0144]

[0145] In the formula, Represents the induction motor in a synchronous rotating coordinate system Axis transient electromotive force;

[0146] Represents the induction motor in a synchronous rotating coordinate system Axis transient electromotive force;

[0147] express The stator-side current component generated by the shaft transient electromotive force;

[0148] express The stator-side current component generated by the shaft transient electromotive force;

[0149] Indicates the reactance of the stator;

[0150] This represents the transient short-circuit reactance of an induction motor.

[0151] This represents the rotor transient time constant when the stator is open-circuited;

[0152] Indicates the reference synchronous angular velocity of the power grid;

[0153] Indicates rotor slip;

[0154] Represents the combined inertial time constant;

[0155] Indicates the rotor's mechanical torque;

[0156] This represents the rotor electromagnetic torque.

[0157] Step 2: Construct an electromechanical equivalent model based on the time-domain simulation model; initialize the electromechanical equivalent model based on the principle of consistent network loss and the principle of capacity weighting to obtain the initialized electromechanical equivalent model.

[0158] 2.1 Construct an electromechanical equivalent model.

[0159] In the time-domain simulation model, the static load, dynamic load, compensation capacitor, and photovoltaic are respectively represented as equivalent components, and all equivalent components are connected to the same virtual bus. The virtual bus is connected to the point of common coupling via an equivalent impedance. The equivalent components include equivalent static load, equivalent dynamic load, equivalent capacitor, and equivalent photovoltaic.

[0160] The equivalent model constructed by the patent is as follows: Figure 9 As shown, the equivalent units in the electromechanical equivalent model mainly include distributed photovoltaic equivalent units, capacitor equivalent units, static load equivalent units, and dynamic load equivalent units. All equivalent units are connected to the virtual bus (Point of Connection, POC) and then connected to the Point of Common Coupling (PCC) via equivalent impedance.

[0161] 2.2 Initialization of the electromechanical equivalent model.

[0162] 2.2.1. Based on the reactive power of each compensation capacitor in the time-domain simulation model, the capacitance value of the equivalent capacitor is calculated.

[0163] The equivalent capacitance of the distribution network, attributed to the virtual bus POC, is obtained based on the reactive power output capacity of all compensation capacitors in the distribution network, as follows:

[0164]

[0165]

[0166] In the formula, This represents the equivalent capacitor reactive power of the distribution network attributed to the virtual bus POC;

[0167] Indicates distribution network node The reactive power of the connected capacitor;

[0168] This represents the equivalent capacitance of the distribution network referred to the virtual bus POC;

[0169] Indicates the system's rated operating frequency;

[0170] This represents the set of all nodes within a power distribution network.

[0171] The voltage of the equivalent virtual bus POC in the distribution network

[0172] 2.2.2 Based on the rated power and voltage characteristic coefficients of each static load in the time-domain simulation model, the equivalent rated power and voltage characteristic coefficients of the equivalent static load are calculated according to the capacity-weighted principle, as follows:

[0173]

[0174]

[0175] In the formula, This represents the rated active power of the equivalent static load at the POC (Point of Contention) of the distribution network;

[0176] This represents the rated reactive power of the equivalent static load at the POC (Point of Contention) of the distribution network;

[0177] The active power proportionality coefficient represents the active power ratio of a constant impedance load.

[0178] The reactive power proportionality coefficient represents the reactive power of a constant impedance load.

[0179] The active power ratio coefficient represents the active power ratio of a constant current load.

[0180] The reactive power proportionality coefficient represents the reactive power of a constant current load.

[0181] The active power ratio coefficient represents the active power ratio of a constant power load.

[0182] The reactive power ratio coefficient represents the reactive power ratio of a constant power load.

[0183] Indicates the first The active power ratio coefficient of constant impedance load at each load node.

[0184] Indicates the first The rated active power of each load node;

[0185] Indicates the first The active power ratio coefficient of constant current load at each load node.

[0186] Indicates the first The reactive power ratio of constant impedance loads at each load node.

[0187] Indicates the first Rated reactive power of each load node;

[0188] Indicates the first The reactive power ratio of constant current loads at each load node.

[0189] 2.2.3. Based on the electrical parameters, mechanical parameters, and rated capacity of each dynamic load in the time-domain simulation model, the equivalent electrical parameters and equivalent mechanical parameters of the equivalent dynamic load are calculated according to the capacity weighting principle.

[0190] The induction motor is the most important component of the dynamic load. The parameters that need to be weighted are the inherent electrical and mechanical parameters in its equivalent circuit model. The formula for calculating a certain parameter using the capacity-weighted method is shown below:

[0191]

[0192] In the formula, Represents a node The parameters of the induction motor are expressed as per-unit values ​​based on its own capacity.

[0193] Per-unit values ​​representing the inherent electrical or mechanical parameters of the induction motor after equivalent values;

[0194] Indicates the first The proportion of the rated capacity of the motor at each node to the total rated capacity of all motors participating in the equivalent value.

[0195] 2.2.4. Based on the rated power of photovoltaic and the active power loss of the network in the time-domain simulation model, the equivalent rated power of the photovoltaic is calculated.

[0196] The photovoltaic capacity in the distribution network is equivalently transferred to the virtual bus POC, as calculated below:

[0197]

[0198] In the formula, This represents the equivalent rated active power of photovoltaic power at the POC node in the equivalent model; This represents the active power loss generated in the distribution network when all photovoltaic power is injected with rated current; Indicates distribution network node The rated active power of photovoltaics.

[0199] Based on the power conservation between PCC and POC, the following conditions must be met in the equivalent distribution network system:

[0200]

[0201] In the formula, This represents the active power flowing from PCC to POC; , , These represent the equivalent impedance of the distribution network, the static load, and the active power consumed by the induction motor, respectively. This represents the active power injected into the equivalent distribution network by the equivalent photovoltaic system;

[0202] This represents the reactive power flowing from PCC to POC; , , These represent the equivalent impedance of the distribution network, the static load, and the reactive power consumed by the induction motor, respectively. , These represent the reactive power injected into the equivalent distribution network by the equivalent capacitor and the equivalent photovoltaic system, respectively.

[0203] 2.2.5. Based on the power conservation relationship between the virtual bus and the point of common coupling, the resistance and reactance values ​​of the equivalent impedance are calculated according to the principle of consistent network loss.

[0204] The equivalent impedance of the distribution network is calculated based on the principle of consistent network loss, as follows:

[0205]

[0206] In the formula, Indicates the resistance of the distribution network; Indicates the reactance of the distribution network; This indicates the voltage at PCC.

[0207] Step 3: Based on the initialized electromechanical equivalent model, the parameter vector to be identified is obtained through transient segmented trajectory sensitivity analysis.

[0208] The specific steps include:

[0209] 3.1 Determine multiple dynamic parameters of the initialized electromechanical equivalent model.

[0210] 3.2 The transient response process of the initialized electromechanical equivalent model is divided into the fault period and the recovery period. By calculating the active power sensitivity, reactive power sensitivity, active power sensitivity and reactive power sensitivity of each dynamic parameter during the fault period, the global comprehensive sensitivity of each dynamic parameter is obtained.

[0211] To clarify the impact of parameters, the electromechanical transient process is divided into a fault period and a recovery period. For example... Figure 10 As shown, Indicates the time when the fault occurred. This indicates the time when the fault is cleared. During the fault period, considering the impact of low-voltage ride-through control of distributed photovoltaic (PV) systems, the distributed PV systems enter a low-voltage ride-through state, increasing reactive power output to support the voltage. At all times, it reaches maximum reactive power output. During the recovery period, distributed photovoltaic power... Always exit low-wear mode.

[0212] The active and reactive power components in the difference between the equivalent model system and the actual system are decoupled and analyzed. The active and reactive power components are then analyzed with respect to the parameters. Sensitivity calculation is as follows:

[0213]

[0214] In the formula, Indicates the selected first One dynamic parameter is used as a perturbation parameter;

[0215] Indicates the first The perturbation of each parameter;

[0216] This represents the set of other dynamic parameters that are not used as perturbation parameters;

[0217] Indicates parameters exist The sensitivity of active and reactive power trajectories at any given time;

[0218] Indicates parameters exist The sensitivity of active and reactive power trajectories at any given time;

[0219] express Real-time details of the active power of the distribution network at the PCC;

[0220] express Real-time details of the reactive power of the distribution network at the PCC;

[0221] Indicates only for the first Apply tiny perturbations to each parameter Afterwards, The active power output of the equivalent model at time step;

[0222] This indicates that only the first... Apply tiny perturbations to each parameter Afterwards, The reactive power output of the equivalent model at any given time.

[0223] Based on the aforementioned transient phase division, in order to quantify the impact of each dynamic parameter on the external characteristics of the distribution network, calculation formulas for the active and reactive power decoupling trajectory sensitivity are established for the fault period and the recovery period, respectively.

[0224] During the outage:

[0225]

[0226] During recovery:

[0227]

[0228] In the formula, Indicates the time when the fault occurred; Indicates the time when the fault is cleared; This is the moment when the PCC response curve recovers to its initial steady-state value after a fault;

[0229] The active power during the fault period is related to the first... Parameters Trajectory sensitivity;

[0230] Represents the reactive power during the fault period with respect to the first Parameters Trajectory sensitivity;

[0231] The active power during the recovery period is related to the first... Parameters Trajectory sensitivity;

[0232] Represents the reactive power during the recovery period with respect to the first... Parameters Trajectory sensitivity.

[0233] 3.3 Select a preset number of dynamic parameters that rank first in global comprehensive sensitivity to form the parameter vector to be identified, and take fixed values ​​for the remaining dynamic parameters.

[0234] In this embodiment, considering both model accuracy and identification efficiency, the active and reactive power trajectory sensitivities of each parameter during the fault and recovery periods are directly added together as the global comprehensive sensitivity of the parameter. The parameters with the highest global comprehensive sensitivity are selected as the parameters to be identified, and the remaining parameters are taken as typical fixed values.

[0235] Step 4: Generate a set of fault scenarios based on the time-domain simulation model.

[0236] For electromechanical equivalent modeling of active power distribution networks with distributed photovoltaic systems, under a single fault scenario, multiple sets of parameters often result in high fitting accuracy of the equivalent model, i.e., parameter non-uniqueness. To address the problem of "different parameters having the same effect" leading to a lack of generalization ability in single fault scenario identification, this paper proposes a multi-scenario validation approach to select the optimal generalization parameters.

[0237] The specific steps include:

[0238] 4.1 At the common connection point of the time-domain simulation model, construct a voltage drop set containing different target drop depths according to the preset voltage step size.

[0239] To effectively stimulate the transient characteristics of distributed photovoltaic and dynamic loads within the active distribution network, a multi-fault scenario set is constructed based on the voltage ride-through range specified in the grid connection standard. At the point of common coupling (PCC), a fixed voltage step size is applied. Construct a set of voltage target drop depths.

[0240]

[0241] In the formula, This represents the set of voltage drop depths at PCC. Indicates the first The per-unit value of the target drop voltage; This represents the minimum voltage threshold for distributed photovoltaic systems to remain connected to the grid without triggering hardware protection to disconnect from the grid; in this embodiment, the value is 0.2 pu. This indicates the total number of scenes generated.

[0242] 4.2 For each target dip depth in the voltage dip set, adjust the short-circuit fault conditions in the external transmission network until the dip voltage at the point of common coupling during the fault reaches the target dip depth. Take the corresponding short-circuit fault conditions as the fault scenario corresponding to the target dip depth. Traverse all target dip depths to obtain a fault scenario set consisting of multiple fault scenarios.

[0243] Specifically, in the benchmark simulation model that includes detailed transmission networks and active distribution networks, for the set The depth of each target's fall Set corresponding short-circuit fault conditions in the external transmission network. This includes adjusting the location of short-circuit faults in the external transmission network and the voltage drop depth, so that the voltage drop at the PCC point of the distribution network reaches the target drop depth during the fault. Iterate through all target fall depths and generate a sequence of values. A set of fault scenarios consisting of external fault conditions.

[0244] Step 5: For each fault scenario in the fault scenario set, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is adopted to assign values ​​to the parameter vector to be identified and optimize it, thereby obtaining the candidate parameter vector for each fault scenario. The candidate parameter vector set is composed of all the candidate parameter vectors.

[0245] For any fault scenario, the optimal equivalent parameter vector, i.e., the candidate parameter vector, is obtained by minimizing the error function. Traverse all Each fault scenario was optimized to form a candidate parameter vector set. The specific steps are as follows:

[0246] 5.1 Objective function.

[0247] Both the adaptive particle swarm optimization algorithm and the Nelder-Mead simplex method in the two-stage hybrid optimization strategy aim to minimize the error function and optimize the vector of parameters to be identified. The error function is used to characterize the difference in power response between the electromechanical equivalent model and the time-domain simulation model at the common connection point.

[0248] Based on the aforementioned sensitivity analysis of decoupling active and reactive power in the segmentation phase, a comprehensive weighted error function for active and reactive power that distinguishes between the fault and recovery periods is constructed as the optimization objective during parameter identification, as shown in the following formula:

[0249]

[0250] In the formula, Represents the parameter vector to be identified The error function, , Indicates the number of parameters to be identified;

[0251] Indicates the weighting coefficient during the fault period;

[0252] The weighting coefficient represents the active power error during a fault.

[0253] The weighting coefficient representing the reactive power error during a fault. ;

[0254] This indicates the active power error during the fault period;

[0255] This indicates the reactive power error during the fault period;

[0256] This represents the weighting coefficient during the recovery period. ;

[0257] The weighting coefficient represents the active power error during the recovery period;

[0258] The weighting coefficient represents the reactive power error during the recovery period. ;

[0259] This indicates the active power error during the recovery period;

[0260] This indicates the reactive power error during the recovery period.

[0261] The weighting coefficients are calculated based on trajectory sensitivity using normalization. The weights during a fault are shown in the following formulas:

[0262]

[0263]

[0264]

[0265] Active and reactive power errors during faults , The following formulas are shown respectively:

[0266]

[0267]

[0268] In the formula, This indicates the total number of data sampling points during the fault period; Represents the sequence number of discrete sampling time points;

[0269] Indicates the parameter vector to be identified Below, the equivalent model during the fault period... Simulated active power values ​​output from each sampling point;

[0270] Indicates the parameter vector to be identified Below, the equivalent model during the fault period... Simulated reactive power values ​​output from each sampling point;

[0271] This indicates the detailed distribution network during the fault period. The active power response value at each sampling point;

[0272] This indicates the detailed distribution network during the fault period. The reactive power response value at each sampling point;

[0273] The active power during the fault period is related to the first... Parameters Trajectory sensitivity;

[0274] Represents the reactive power during the fault period with respect to the first Parameters Trajectory sensitivity;

[0275] The active power during the recovery period is related to the first... Parameters Trajectory sensitivity;

[0276] Represents the reactive power during the recovery period with respect to the first... Parameters Trajectory sensitivity.

[0277] Active and reactive power errors during recovery , The following formulas are shown respectively:

[0278]

[0279]

[0280] In the formula, This indicates the total number of data sampling points during the recovery period;

[0281] Indicates the parameter vector to be identified Below, the equivalent model during the recovery period... Simulated active power values ​​output from each sampling point;

[0282] Indicates the parameter vector to be identified Below, the equivalent model during the recovery period... Simulated reactive power values ​​output from each sampling point;

[0283] This indicates the detailed distribution network during the restoration period. The active power response value at each sampling point;

[0284] This indicates the detailed distribution network during the restoration period. The reactive power response value of each sampling point.

[0285] A two-stage hybrid optimization strategy is adopted for parameter identification. First, an improved particle swarm optimization algorithm is used to obtain preliminary solutions for the parameters to be identified. Then, these initial values ​​are substituted into the Nelder-Mead simplex method for local fine-tuning to obtain accurate solutions, as detailed below:

[0286] 5.2 Adaptive Particle Swarm Optimization Algorithm

[0287] The first stage identifies the preliminary solution of the parameters. To address the problem that traditional Particle Swarm Optimization (PSO) algorithms are prone to getting trapped in local optima and have low convergence efficiency in multi-parameter, high-dimensional optimization, the Adaptive Particle Swarm Optimization (APSO) algorithm improves the particle velocity update equation, inertia weight, and learning factor.

[0288] In APSO, the first The parameter of the first The particle in the first The velocity update equation for the next iteration is:

[0289]

[0290] In the formula, Indicates the first The parameter of the first The particle in the first Flight speed at the next iteration;

[0291] Indicates the first The parameter of the first The particle in the first The position at the next iteration;

[0292] Indicates the first Inertia weights in the next iteration;

[0293] Indicates the first Individual learning factor at the next iteration;

[0294] They represent the first Social learning factor at the next iteration;

[0295] This represents the best historical position of an individual particle.

[0296] Indicates the globally optimal position of the population;

[0297] The first random number represents a uniformly distributed random number;

[0298] This represents a second random number from a uniform distribution. .

[0299] To ensure the algorithm has strong global exploration capability in the early stages of iteration and high local convergence accuracy in the later stages, an adaptive inertia weight that decreases linearly with the number of iterations is used. The calculation formula is as follows:

[0300]

[0301] In the formula, , These represent the maximum and minimum inertia weights, respectively, which are set to 0.9 and 0.4 in this embodiment. Indicates the current iteration number; This indicates the maximum number of iterations.

[0302] To overcome the premature convergence of the population caused by the fixed learning factor in traditional PSO, APSO introduces a time-varying learning factor. In the early stages of iteration, a large individual and small social learning factor is used to enhance independent exploration; as iteration progresses, the two factors decrease linearly and increase linearly, respectively, thereby accelerating the convergence of particles to the global optimum. The specific formula is as follows:

[0303]

[0304] In the formula, This represents the initial value of the individual learning factor, which is 2.5.

[0305] This represents the initial value of the social learning factor, which is 0.5.

[0306] This represents the final value of the individual learning factor, which is 0.5.

[0307] This represents the final value of the social learning factor, which is 2.5.

[0308] To prevent particles from escaping the solution space, their velocities need to be limited. Considering the differences in the dimensions and orders of magnitude of the multiple parameters, a maximum allowable velocity is independently set for each parameter. The calculation formula is as follows:

[0309]

[0310] In the formula, , They represent the first The upper and lower limits of the search space for each parameter; This represents the speed proportionality coefficient, in this embodiment... .

[0311] 5.3 Nelder-Mead Simplex Method

[0312] Using the initial solution obtained by the APSO algorithm as the initial value, the Nelder-Mead simplex method is employed for local refinement search. Before each iteration, the error values ​​of each vertex are calculated and sorted. The algorithm continuously updates the worst-case vertex through four operations: reflection, expansion, contraction, and shrinkage. The specific iterative steps and calculation formulas are as follows:

[0313] 5.3.1: Calculate the values ​​of all vertices except the worst vertex. The geometric centroid of the better vertex .

[0314]

[0315] 5.3.2: Remove the worst vertex Along the center of gravity Perform reflection and calculate the reflection point. .

[0316]

[0317] like This indicates that the reflection point is good but not optimal, so we can directly use... replace .

[0318] 5.3.3: If the reflection point performs exceptionally well, that is... Then, extend along this direction further and calculate the expansion point.

[0319]

[0320] like ,use replace Otherwise use replace.

[0321] 5.3.4: If the reflection operation fails to achieve a result superior to... If the solution is not found, the search range needs to be narrowed down by performing a shrinking operation.

[0322] when When calculating the external contraction point

[0323]

[0324] when When calculating the internal contraction point

[0325]

[0326] 5.3.5: If the latest point generated by all the above operations fails to improve the error value, then the current best vertex shall be used. Indicates the baseline, scaled down proportionally.

[0327]

[0328] In the formula, Represents the parameter vector to be identified The error function, , Indicates the number of parameters to be identified;

[0329] This represents the reflection coefficient, which is set to 1 in this embodiment;

[0330] This represents the coefficient of thermal expansion, which is set to 2 in this embodiment;

[0331] This represents the shrinkage coefficient, which is 0.5 in this embodiment;

[0332] This represents the reduction factor, which is 0.5 in this embodiment.

[0333] Step 6: Substitute the candidate parameter vector set into the fault scenario set to perform electromechanical transient simulation, and calculate the multi-scenario generalization error of each candidate parameter vector set.

[0334] The specific steps include:

[0335] 6.1 For each set of candidate parameter vectors in the candidate parameter vector set, substitute them into each fault scenario in the fault scenario set to perform electromechanical transient simulation, obtain the active power simulation value and reactive power simulation value of the electromechanical equivalent model at the common connection point under each fault scenario; obtain the active power response value and reactive power response value of the time domain simulation model under the fault scenario.

[0336] candidate parameter vector set Each set of candidate parameter vectors Substitute all of them Electromechanical transient verification is performed in the fault scenarios of an equivalent simulation system.

[0337] Electromechanical transient simulation is performed under the k-th fault scenario. The active power response value and reactive power response value transmitted between the time-domain simulation model and the external power grid at the PCC are recorded as reference targets for parameter identification.

[0338]

[0339] In the formula, Indicates the first Under a fault scenario, the active power response value transmitted between the time-domain simulation model and the external power grid at the PCC.

[0340] Indicates the first Under a fault scenario, the reactive power response value transmitted between the time-domain simulation model and the external power grid at the PCC.

[0341] This represents the total number of sampling points within the observation period.

[0342] Retaining the aforementioned transmission network topology, the time-domain simulation model is replaced with the electromechanical equivalent model to be identified. The candidate parameter vector set... Each set of candidate parameter vectors Substitute the values ​​into the electromechanical equivalent model. Strictly reproduce the aforementioned external fault conditions within the external transmission network of the electromechanical equivalent model. Under the external fault excitation, the simulated active power and reactive power transmitted between the electromechanical equivalent model and the external power grid at the PCC are obtained.

[0343] 6.2. Based on the simulated active power value, simulated reactive power value, active power response value, and reactive power response value, calculate the root mean square error of each candidate parameter vector under each fault scenario.

[0344] The specific formula for single-scene error is as follows:

[0345]

[0346] In the formula, Indicates the first In the first failure scenario, use the first candidate parameter vectors The calculated root mean square error;

[0347] Indicates the first The total number of sampling points for each fault scenario;

[0348] The time-domain simulation model is represented in the first... In the first fault scenario, the first The active power response value at the common connection point PCC at each sampling time;

[0349] The time-domain simulation model is represented in the first... In the first fault scenario, the first The reactive power response value at the common connection point PCC at each sampling time;

[0350] Indicates the first candidate parameter vectors After substituting into the electromechanical equivalent model, the first In the first fault scenario, the first Simulated active power values ​​at the common connection point PCC at each sampling time;

[0351] Indicates the first candidate parameter vectors After substituting into the electromechanical equivalent model, the first In the first fault scenario, the first Simulated reactive power values ​​at the common connection point PCC at each sampling time.

[0352] 6.3. The root mean square error of each candidate parameter vector in each fault scenario is weighted and summed to obtain the multi-scenario generalization error of each candidate parameter vector.

[0353] The formula for the generalization error in multiple scenarios is as follows:

[0354]

[0355] In the formula, Indicates the first candidate parameter vectors Multi-scenario generalization error; Indicates the first The weight of each fault scenario in the evaluation of all fault scenarios is represented by the sum of the weights of all scenarios, where 1 represents the sum of the weights of all scenarios.

[0356] Step 7: Select the candidate parameter vector with the smallest generalization error in multiple scenarios as the final parameter vector of the electromechanical equivalent model, so as to realize the identification of equivalent parameters for minimizing the generalization error in multi-disturbance scenarios.

[0357] In summary, the technical improvements of this invention are as follows:

[0358] First, considering the difference in dynamic response between the fault period and the recovery period of the active distribution network, the electromechanical transient response is divided into the fault period and the recovery period. The trajectory sensitivity of active power and reactive power at the point of common coupling to each dynamic parameter is decoupled and calculated. The active power sensitivity, reactive power sensitivity, active power sensitivity, and reactive power sensitivity during the fault period, the recovery period, and the recovery period are obtained by integration respectively. Based on this, the dominant parameters of different transient stages are screened out, effectively reducing the dimensionality of the parameters to be identified.

[0359] Secondly, a two-stage hybrid optimization strategy based on the combination of an improved particle swarm optimization algorithm and the Nelder-Mead simplex method was constructed. In the first stage, an improved particle swarm optimization algorithm with adaptive inertia weights and time-varying learning factors was used for global optimization to obtain an initial solution. In the second stage, the initial solution was used as the starting point, and the Nelder-Mead simplex method was used to perform local precise optimization through reflection, expansion, contraction, and reduction operations, which takes into account both global exploration capability and local convergence accuracy.

[0360] Finally, by changing the location and intensity of disturbances in the external power grid, a set of fault scenarios containing multiple fault scenarios is constructed. The candidate parameter vectors independently identified under each fault scenario are used to form a set of candidate parameter vectors. Then, all fault scenarios are substituted into the electromechanical transient simulation to calculate the multi-scenario generalization error of each set of candidate parameter vectors. The final parameter vector is selected based on the principle of minimizing the multi-scenario generalization error, thus solving the problems of different parameters having the same effect and overfitting caused by single-scenario identification.

[0361] The beneficial effects of this invention are as follows:

[0362] First, it significantly reduces computational dimensionality and time consumption. By screening key parameters through transient segmented trajectory sensitivity analysis and fixing low-sensitivity parameters to typical values, the number of parameters to be identified is greatly reduced, thus improving the computational efficiency of electromechanical transient simulation.

[0363] Secondly, it balances global accuracy and high precision in identification. By combining the global optimization capability of the improved particle swarm optimization algorithm with the local fine-grained search capability of the Nelder-Mead simplex method, it avoids the problem of single algorithms easily getting trapped in local optima, thus improving the accuracy of parameter identification for equivalent models.

[0364] Finally, the generalization ability of the equivalent model was improved. Through multi-scenario verification, the optimal parameters were selected based on minimizing the multi-scenario generalization error, reducing overfitting and the problem of different parameters having the same effect caused by single-scenario identification, so that the electromechanical equivalent model can adapt to external disturbances of different locations and intensities.

[0365] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0366] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0367] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0368] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0369] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An isoparameter identification method for minimizing generalization error in multi-perturbation scenarios, characterized by: include: Obtain topology and operational data of active power distribution networks containing high-penetration distributed photovoltaic power. Based on the topology and operational data, a time-domain simulation model of the power distribution network is constructed; Based on the time-domain simulation model, an electromechanical equivalent model is constructed; the electromechanical equivalent model is initialized based on the network loss consistency principle and the capacity weighting principle to obtain the initialized electromechanical equivalent model; Based on the initialized electromechanical equivalent model, the parameter vector to be identified is obtained through transient segmented trajectory sensitivity analysis. Based on the time-domain simulation model, a set of fault scenarios is generated. For each fault scenario in the set of fault scenarios, a two-stage hybrid optimization strategy combining the improved particle swarm optimization algorithm and the Nelder-Mead simplex method is used to assign and optimize the parameter vector to be identified, thereby obtaining the candidate parameter vector under each fault scenario. The candidate parameter vector set is composed of all the candidate parameter vectors. Substitute the candidate parameter vector set into the fault scenario set to perform electromechanical transient simulation, and calculate the multi-scenario generalization error of each candidate parameter vector set. The candidate parameter vector with the smallest generalization error in the multi-scenario model is selected as the final parameter vector of the electromechanical equivalent model, thereby realizing the identification of equivalent parameters for minimizing the generalization error in multi-disturbance scenarios.

2. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The construction of the time-domain simulation model of the distribution network includes: Based on the topology, determine the access node locations of static loads, dynamic loads, compensation capacitors, and photovoltaics, as well as the electrical connection relationships between distribution network lines and transformers; Based on the aforementioned operational data, mathematical models for photovoltaic, static load, and dynamic load are established respectively. Based on the mathematical models of photovoltaic, static load and dynamic load, interconnection is performed according to the location and electrical connection relationship of the access node, and the reactive power compensation capacity parameters of the compensation capacitor and the impedance parameters of the distribution network lines and transformers are connected at the corresponding access node to construct a time-domain simulation model of the distribution network.

3. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 2, characterized in that, The mathematical model of the photovoltaic system includes a mathematical model of type A photovoltaic system and a mathematical model of type B distributed photovoltaic system. The mathematical model of type A photovoltaic system adopts fault ride-through control logic that includes calculation of active current command and reactive current command during low voltage ride-through. The mathematical model of type B distributed photovoltaic system adopts a blocking control logic that includes power attenuation logic under blocking control during fault and power linear recovery logic after fault clearance. The mathematical model of the static load adopts a polynomial static load model that includes constant impedance components, constant current components, and constant power components. The mathematical model of the dynamic load uses a third-order electromechanical transient model to characterize the dynamic response of the induction motor group.

4. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The construction of the electromechanical equivalent model includes: In the time-domain simulation model, the static load, dynamic load, compensation capacitor, and photovoltaic are respectively represented as equivalent components, and all equivalent components are connected to the same virtual bus. The virtual bus is connected to the point of common connection via an equivalent impedance. The equivalent components include equivalent static load, equivalent dynamic load, equivalent capacitor, and equivalent photovoltaic.

5. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The obtained initialized electromechanical equivalent model includes: Based on the reactive power of each compensation capacitor in the time-domain simulation model, the capacitance value of the equivalent capacitor is calculated. Based on the rated power and voltage characteristic coefficients of each static load in the time-domain simulation model, the equivalent rated power and voltage characteristic coefficients of the equivalent static load are calculated according to the capacity weighting principle. Based on the electrical parameters, mechanical parameters, and rated capacity of each dynamic load in the time-domain simulation model, the equivalent electrical parameters and equivalent mechanical parameters of the equivalent dynamic load are calculated according to the capacity weighting principle. Based on the rated power of the photovoltaic and the active power loss of the network in the time-domain simulation model, the equivalent rated power of the equivalent photovoltaic is calculated. Based on the power conservation relationship between the virtual bus and the common connection point, the resistance and reactance values ​​of the equivalent impedance are calculated according to the principle of consistent network loss, thus completing the initialization of the electromechanical equivalent model.

6. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The filtering process yields a vector of parameters to be identified, including: Determine multiple dynamic parameters of the initialized electromechanical equivalent model; The transient response process of the initialized electromechanical equivalent model is divided into a fault period and a recovery period. By calculating the active power sensitivity, reactive power sensitivity, active power sensitivity, and reactive power sensitivity of each dynamic parameter during the fault period, the global comprehensive sensitivity of each dynamic parameter is obtained. A preset number of dynamic parameters ranked first in global comprehensive sensitivity are selected to form the parameter vector to be identified, while the remaining dynamic parameters are taken as fixed values.

7. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The generated fault scenario set includes: At the common connection point of the time-domain simulation model, a voltage drop set containing different target drop depths is constructed according to a preset voltage step size; For each target dip depth in the voltage dip set, the short-circuit fault conditions in the external transmission network are adjusted until the dip voltage at the point of common coupling during the fault reaches the target dip depth, and the corresponding short-circuit fault conditions are taken as the fault scenarios corresponding to the target dip depth. By iterating through all target drop depths, a set of failure scenarios consisting of multiple failure scenarios is obtained.

8. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The two-stage hybrid optimization strategy includes: For each fault scenario in the fault scenario set, an adaptive particle swarm optimization algorithm is used for global search to obtain a preliminary solution for the parameter vector to be identified. Using the preliminary solution of the parameter vector to be identified as the initial point, the Nelder-Mead simplex method is used for local search to obtain the candidate parameter vector; The adaptive particle swarm optimization algorithm and the Nelder-Mead simplex method both aim to minimize the error function to optimize the vector of parameters to be identified. The error function is used to characterize the difference in power response between the electromechanical equivalent model and the time-domain simulation model at the common connection point.

9. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 8, characterized in that, The expression for the error function is as follows: ； In the formula, Represents the parameter vector to be identified The error function, , Indicates the number of parameters to be identified; Indicates the weighting coefficient during the fault period; The weighting coefficient represents the active power error during a fault. The weighting coefficient represents the reactive power error during a fault. This indicates the active power error during the fault period; This indicates the reactive power error during the fault period; Indicates the weighting coefficient during the recovery period; The weighting coefficient represents the active power error during the recovery period; The weighting coefficient represents the reactive power error during the recovery period; This indicates the active power error during the recovery period; This indicates the reactive power error during the recovery period.

10. The method for identifying equivalent parameters for minimizing generalization error in multi-perturbation scenarios according to claim 1, characterized in that, The calculation of the multi-scenario generalization error for each set of candidate parameter vectors includes: For each set of candidate parameter vectors in the candidate parameter vector set, substitute them into each fault scenario in the fault scenario set to perform electromechanical transient simulation, and obtain the active power simulation value and reactive power simulation value of the electromechanical equivalent model at the common connection point under each fault scenario; Obtain the active power response value and reactive power response value of the time-domain simulation model under the fault scenario; Based on the simulated active power value, simulated reactive power value, active power response value, and reactive power response value, calculate the root mean square error of each candidate parameter vector under each fault scenario. The root mean square error of each candidate parameter vector under each fault scenario is weighted and summed to obtain the multi-scenario generalization error of each candidate parameter vector.

Images