Generalized comprehensive load modeling method for distribution system based on Bayesian estimation theory

Abstract

The present invention discloses a generalized comprehensive load modeling method for a distribution system based on Bayesian estimation theory, which relates to the technical field of distribution system modeling, and includes the following steps: obtaining a photovoltaic grid-connected sub-model, an energy storage grid-connected sub-model, an induction motor sub-model, and a static load sub-model included in the generalized comprehensive load of the distribution system; using RBF neural network to model the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model, and using parameter identification method to model the induction motor sub-model and the static load sub-model; based on the modeling results, using Bayesian estimation theory to calculate the probability of each sub-model appearing in the generalized comprehensive load, obtaining the output result of the generalized comprehensive load model of the distribution system, and completing the generalized comprehensive load modeling of the distribution system based on Bayesian estimation theory. The present invention solves the problem that traditional methods cannot handle the complex dynamic responses brought by photovoltaic and energy storage systems in the distribution system.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution system modeling, and in particular to a generalized integrated load modeling method for a power distribution system based on Bayesian estimation theory. Background Art

[0002] Generalized load modeling refers to the process of modeling load nodes that include distributed generation resources (such as wind power, photovoltaics, small hydropower, etc.) in addition to traditional loads (such as residential electricity, industrial electricity, etc.) in the power system. This modeling method not only takes into account the consumption characteristics of the load, but also takes into account the power generation characteristics that the load nodes may have, so as to more comprehensively reflect the actual situation of the power system under complex operating conditions. The traditional comprehensive load model is unable to cope with the complex dynamic responses brought about by photovoltaic and energy storage systems in the distribution system, because the access of these new energy sources makes the demand forecasting and management of the power system more sophisticated and complex. Therefore, power system load modeling must keep pace with the times and incorporate photovoltaic power generation systems and energy storage systems into a more extensive and dynamic load model to meet the challenges of the times. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the generalized integrated load modeling method for distribution system based on Bayesian estimation theory provided by the present invention solves the problem that traditional methods cannot handle the complex dynamic responses brought about by photovoltaic and energy storage systems in distribution systems.

[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a generalized integrated load modeling method of a distribution system based on Bayesian estimation theory, comprising the following steps:

[0005] S1: Obtain the generalized comprehensive load of the distribution system, including the photovoltaic grid-connected sub-model, energy storage grid-connected sub-model, induction motor sub-model and static load sub-model;

[0006] S2: RBF neural network is used to model the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model, and parameter identification method is used to model the induction motor sub-model and the static load sub-model;

[0007] S3: Based on the modeling results, the Bayesian estimation theory is used to calculate the probability of each sub-model appearing in the generalized comprehensive load, and the output results of the generalized comprehensive load model of the distribution system are obtained, completing the generalized comprehensive load modeling of the distribution system based on the Bayesian estimation theory.

[0008] Further, the photovoltaic grid-connected sub-model includes a PV array, a drive Boost circuit, a three-phase voltage source inverter, an LCL filter and a power grid connected in sequence;

[0009] The energy storage grid-connected sub-model includes an energy storage battery, an inverter circuit, an L-type filter circuit and a power grid connected in sequence;

[0010] The induction motor sub-model is represented by a third-order induction motor model. The rotor voltage equation of the third-order induction motor model is:

[0011]

[0012]

[0013]

[0014] in, and They are respectively the motor's transient state Axis and Axis potential, For time, is the transient open circuit time constant, is the rotor open-circuit reactance, is the transient reactance of the induction motor, and The stator winding current is Axis and The weight on the axis, is the synchronous angular velocity of the rotor, is the actual angular velocity of the rotor, and is the stator winding leakage reactance, is the magnetizing reactance, is the rotor winding resistance;

[0015] The stator current equation of the third-order induction motor model is:

[0016]

[0017] in, is the stator winding resistance, and The stator winding terminal voltage is Axis and Components on the axis;

[0018] The rotor motion equation of the third-order induction motor model is:

[0019]

[0020]

[0021] in, is the motor rotor inertia time constant, is the slip rate of the motor, is the electromagnetic torque of the motor, is the mechanical torque of the motor, and They are the stator current in the transient process. Axis and The weight of the axis, is the initial load rate of the induction motor, and is the mechanical torque characteristic parameter;

[0022] The static load sub-model is expressed using the ZIP load model as follows:

[0023]

[0024] in, and are the active power and reactive power consumed by the ZIP load model, and They are The active power and reactive power consumed by the load model, is the actual operating voltage, is the preset reference voltage, , and They represent the active contribution of the load respectively, , and They represent the reactive contribution of the load respectively.

[0025] Furthermore, in S2, the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model are both modeled using RBF neural networks, and the input layers of the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model both receive voltage and frequency change signals from the power grid, and the output layers both output dynamic active power and reactive power signals of the photovoltaic grid-connected power generation system.

[0026] Furthermore, in said S2, both the induction motor sub-model and the static load sub-model are modeled by using a parameter identification method, wherein the modeling of the induction motor sub-model by using a parameter identification method comprises the following sub-steps:

[0027] S2-1: Calculate the initial voltage, initial active power, initial reactive power, initial power of the induction motor and initial power of the ZIP load model on the load bus, and initialize the intermediate parameters of the induction motor and the ZIP load model;

[0028] S2-2: Based on the initialization results, the fourth-order Runge-Kutta method is used to solve the induction motor differential equation to obtain the induction motor state variables:

[0029] S2-3: Based on the state variables of the induction motor, the power consumption of the induction motor is solved by using the transient potential of the induction motor, and the criterion function is calculated according to the power consumption of the induction motor;

[0030] S2-4: Use the simulated evolution method to identify the model parameters. If the criterion function is greater than the threshold, return to step S2-2, otherwise complete the parameter identification.

[0031] Furthermore, the calculation formula of the induction motor state variable in S2-2 is:

[0032]

[0033] in, and Induction motor in transient process Axis and The electric potential of the axis, is the angular velocity of the rotor, is the number of estimates in the Runge-Kutta method, is the time interval in the Runge-Kutta method, , , , , , , , , , , and represents each partial differential component;

[0034]

[0035]

[0036]

[0037] .

[0038] Furthermore, the criterion function in S2-3 The calculation formula is:

[0039]

[0040] in, is the data sampling length, and are the measured active power and reactive power respectively, and The induction motor sub-model responds to active power and reactive power, respectively.

[0041] Furthermore, S3 includes the following sub-steps:

[0042] S3-1: using the voltage measurement value, power measurement value and voltage change at the current moment in each established sub-model as input information of each sub-model, and using each sub-model to obtain the output power value according to the input information;

[0043] S3-2: Calculate the probability of each submodule at the next moment based on the output error of the output power value of each submodel and the probability at the current moment using the Bayesian estimation theory;

[0044] S3-3: Weight the probability of each submodule at the next moment in proportion to obtain the output result of the generalized comprehensive load model of the distribution system.

[0045] Furthermore, the probability of each submodule at the next moment in S3-2 is:

[0046]

[0047]

[0048] in, For sub-model In the The posterior probability of matching in the iteration, For sub-model In the The prior probability of matching in the iteration, For sub-model In the The prior probability of matching in the iteration, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The output covariance diagonal matrix, For sub-model The output covariance diagonal matrix, is the total number of sub-models, is the critical value set, the superscript is the transpose of the matrix.

[0049] The beneficial effects of the present invention are:

[0050] 1. The present invention establishes four representative sub-models, namely, motor load, static load, photovoltaic system and energy storage system. The motor and static load are modeled using parameter identification method, and the photovoltaic and energy storage systems are modeled using RBF neural network. Each sub-model is independent of each other and is not significantly restricted by the comprehensive load composition, so the modeling accuracy is relatively high.

[0051] 2. The present invention applies Bayesian estimation theory to load modeling. It integrates the characteristics of each sub-model and can accurately analyze the components of the load, which is conducive to improving the versatility of neural network load modeling.

[0052] 3. The present invention adopts multiple models for load modeling, which can describe the dynamic process of generalized comprehensive load. Compared with the traditional modeling method, it can better describe the dynamic change characteristics of generalized comprehensive load. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of the generalized integrated load modeling method for distribution systems based on Bayesian estimation theory.

[0054] Figure 2 Schematic diagram of the photovoltaic grid-connected sub-model topology and its control.

[0055] Figure 3 This is a schematic diagram of the energy storage grid-connected sub-model topology and its control.

[0056] Figure 4 This is a typical structural diagram of the photovoltaic grid-connected sub-model.

[0057] Figure 5 This is a typical structural diagram of the energy storage grid-connected sub-model.

[0058] Figure 6 This is the schematic diagram of static load sub-model parameter identification.

[0059] Figure 7 Diagram of the multi-model generalized load modeling framework.

[0060] Figure 8 This is the IEEE14-node generalized load access topology diagram.

[0061] Fig. 9 This is the dynamic response change diagram of the active power of the generalized comprehensive load when the voltage drops by 20%.

[0062] Fig.10 This is the dynamic response change diagram of reactive power of generalized comprehensive load when the voltage drops by 20%. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0064] like Figure 1 As shown, a generalized integrated load modeling method for distribution system based on Bayesian estimation theory includes the following steps:

[0065] S1: Obtain the generalized comprehensive load of the distribution system, including the photovoltaic grid-connected sub-model, energy storage grid-connected sub-model, induction motor sub-model and static load sub-model;

[0066] S2: RBF neural network is used to model the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model, and parameter identification method is used to model the induction motor sub-model and the static load sub-model;

[0067] S3: Based on the modeling results, the Bayesian estimation theory is used to calculate the probability of each sub-model appearing in the generalized comprehensive load, and the output results of the generalized comprehensive load model of the distribution system are obtained, completing the generalized comprehensive load modeling of the distribution system based on the Bayesian estimation theory.

[0068] The photovoltaic grid-connected sub-model includes a PV array, a driving Boost circuit, a three-phase voltage source inverter, an LCL filter and a power grid connected in sequence;

[0069] The intensity and temperature of solar radiation will significantly affect the output power of the PV array, but the light and temperature in the natural environment will not change suddenly, and the response time of the PV array to light and temperature changes is relatively long, often up to tens of seconds, while the transient process of voltage and frequency fluctuations in the power system is generally in the second level. Therefore, the influence of light and temperature changes can be ignored when analyzing the transient characteristics of the photovoltaic grid-connected system in the power system. Based on the Matlab / Simulink simulation toolbox, a photovoltaic system grid-connected model is built, such as Figure 2 As shown in the figure, this is used as the basis for studying its external characteristics. The model includes PV array, drive boost circuit, three-phase voltage source inverter and its control circuit, LCL filter and power grid. The drive boost circuit is designed to increase the voltage level on the DC side of the inverter. In the circuit, a diode is configured on the DC bus bar connected in series, and its main function is to prevent the current from flowing in the reverse direction. In addition, the capacitor equipped on the DC bus bar undertakes the key task of maintaining the stability of the DC voltage.

[0070] exist Figure 2 middle, u pv and i pv are the operating voltage and output current of the photovoltaic array respectively; and They are the input and output capacitors driving the Boost circuit respectively; L1C f L2 forms a third-order filter; R 1 and R2 are the internal resistances of L1 and L2 respectively; u g.abc is the grid voltage; , and is the inverter output current, , and is the inverter output voltage; and They are Axis and The voltage output after the inner loop of the axis current is decoupled. and The grid voltage is Axis and Axis component; is the rotation angle of the grid voltage in the dq coordinate system, which defines the offset of the q axis relative to the a phase axis. , and is the three-phase voltage of the power grid.

[0071] The energy storage grid-connected sub-model includes an energy storage battery, an inverter circuit, an L-type filter circuit and a power grid connected in sequence;

[0072] In the MATLAB / Simulink environment, the present invention constructs a simulation model for simulating a battery energy storage system (BESS). This model is based on The decoupling control technology of the synchronous rotating coordinate system adopts a dual closed-loop control architecture in its core control structure to ensure the stability and efficiency of the system. Through the simulation platform, we can test and optimize the performance of BESS. The outer loop is the power loop, which is responsible for monitoring the power of the grid connection point and comparing it with the preset value. The PI controller generates a Axis and The inner loop is the current loop, which monitors the grid-connected current and compares it with the set value. The error signal processed by the PI controller is then decoupled and output , The voltage command of the axis is finally converted and executed through the pulse width modulation (PWM) signal of the inverter. The architecture and control process of BESS are as follows: Figure 3 shown.

[0073] in, U dc Represents the terminal voltage of the battery pack. , The axis components are expressed as i d and i q ,and i* d and i * q The power outer loop is based on the set power reference value P * and Q * The calculated current target value. , Axis Component and , then it corresponds to the voltage output after the current inner loop is decoupled u * d and u * q The system goal is active power P * and reactive power Q * control, , and for , and Three-phase grid current.

[0074] The induction motor sub-model is represented by a third-order induction motor model. The rotor voltage equation of the third-order induction motor model is:

[0075]

[0076]

[0077]

[0078] in, and They are respectively the motor's transient state Axis and Axis potential, For time, is the transient open circuit time constant, which is used to describe the response speed of the motor from short circuit to open circuit state. It is the rotor open-circuit reactance, which affects the dynamic response of the motor. is the transient reactance of the induction motor, which is related to the instantaneous state of the motor. and The stator winding current is Axis and The weight on the axis, is the synchronous angular velocity of the rotor, is the actual angular velocity of the rotor, and is the stator winding leakage reactance, which reflects the damping effect of AC current flowing in the winding. is the excitation reactance, which is related to the magnetic field strength and structure of the motor. It is the rotor winding resistance, which affects the energy consumption and heat generation of the motor;

[0079] The stator current equation of the third-order induction motor model is:

[0080]

[0081] in, is the stator winding resistance, and The stator winding terminal voltage is Axis and Components on the axis;

[0082] The rotor motion equation of the third-order induction motor model is:

[0083]

[0084]

[0085] in, is the motor rotor inertia time constant, which reflects the influence of the motor rotor rotational inertia on the speed change and is related to the motor size and mass. is the slip rate of the motor, which is the ratio of the difference between the actual speed of the motor and the synchronous speed to the synchronous speed. The electromagnetic torque of the motor is the torque generated by the electromagnetic force of the motor and is the main source of power inside the motor. is the mechanical torque of the motor, which depends on the load demand and motor design, and They are the stator current in the transient process. Axis and The weight of the axis, The initial load rate of the induction motor indicates the load condition when the motor starts running and has an impact on the starting performance. and It is a mechanical torque characteristic parameter, usually used to describe the mechanical characteristics of the motor, such as linearity or nonlinearity, and reflects the load capacity of the motor at different speeds;

[0086] The third-order model is more comprehensive. It not only considers the electromagnetic transient response of the rotor, but also includes the impact of slip changes on impedance. Therefore, it is more accurate in capturing the transient dynamic characteristics of the power system, and is therefore more widely used in actual simulation calculations. This distinction helps ensure the accuracy of the model and reduce repeated descriptions.

[0087] The static load sub-model is expressed using the ZIP load model as follows:

[0088]

[0089] in, and are the active power and reactive power consumed by the ZIP load model, and They are The active power and reactive power consumed by the load model, is the actual operating voltage, is the preset reference voltage, , and They represent the active contribution of the load respectively, , and They represent the reactive contribution of the load respectively.

[0090] The static load in the Classic Load Model (CLM) is usually described by a polynomial (ZIP) model or a constant impedance model. In the analysis of this scheme, the contribution of each type of load is expressed by , and (satisfy a p + b p + c p =1) and , and (same a q + b q + c q =1) to quantify their proportions in the total load. U and U 0 represents the actual operating voltage and the preset reference voltage respectively. U When the ZIP load model consumes active power P s and reactive power Q s Compared with the baseline condition ( U 0) and The difference reflects the characteristics of load changing with voltage.

[0091] In S2, the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model are both modeled using RBF neural networks. The input layers of the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model both receive voltage and frequency change signals from the power grid, and the output layers both output dynamic active power and reactive power signals of the photovoltaic grid-connected power generation system.

[0092] The single-stage photovoltaic grid-connected power generation system is mainly composed of photovoltaic cells, front-stage Boost circuit, rear-stage inverter circuit, and LCL filter circuit. Its typical structure is as follows: Figure 4 As shown. Among them, and is the grid-connected current and voltage of the photovoltaic cell, is the voltage of the photovoltaic cell after being boosted by the previous boost circuit. , and is the three-phase voltage of the power grid. The maximum power tracking control method adopted by the present invention is the conductance increment method. This model simulates the dynamic characteristics of the photovoltaic grid-connected system under the premise that the photovoltaic array always provides the maximum available power under the current environmental conditions. This is a common assumption for modeling the transient characteristics of the photovoltaic model.

[0093] The present invention adopts RBF neural network model to describe photovoltaic grid-connected power generation system. The input layer of the model receives signals of voltage and frequency change from the power grid, and the output layer outputs dynamic active power and reactive power signals of the photovoltaic grid-connected power generation system. The neural network model includes three commonly used control strategies for photovoltaic grid-connected systems: Vf control, PQ control and Droop control.

[0094] The unipolar energy storage grid-connected system is mainly composed of energy storage batteries, inverter circuits and L-type filter circuits. Its typical structure is as follows: Figure 5 As shown, the energy storage grid-connected control strategy adopted by the present invention is active-reactive power control.

[0095] The present invention adopts the RBF neural network modeling method to describe the energy storage grid-connected system, adopts the active-reactive power control strategy, collects the power data of the energy storage grid-connected model under different degrees of voltage drop as the training data set of the RBF neural network, the input layer of the model receives the voltage change signal from the power grid, and the output layer outputs the dynamic active power and reactive power signals of the photovoltaic grid-connected power generation system.

[0096] A modeling study is conducted on the generalized comprehensive load in the distribution system. The traditional motor load is represented by the third-order induction motor model, and the static load is represented by the polynomial (ZIP) model. The third-order induction motor model and the ZIP model are traditional load models. The present invention adopts a simulation evolution method to identify load modeling parameters.

[0097] With the advancement of computer technology in the field of data processing, the theory of system identification has been deepened and widely used. It involves three key links: first, collecting the input and output data of the system, second, building a mathematical model, and finally finding a set of optimal parameters based on these data through an optimization algorithm. The goal is to make the model simulate the actual system as accurately as possible, and this goal is achieved by minimizing the error between the model output and the actual result. The core strategy of parameter identification is to obtain data during the power system fault with the help of fault detection equipment. Its purpose is to adjust the parameters of the preset model through the optimization algorithm so that it can accurately reflect the behavior of the actual power grid load. For example, Figure 6 As shown, the measured voltage collected by the substation U As input, a specific model is used to predict the power response P l and Q l Then, an objective function is set to evaluate the model's fitness, i.e., the fitness value J In this process, the optimization algorithm repeatedly adjusts the parameters until the fitness value reaches the minimum, thus maximally approaching the actual load characteristics.

[0098] The induction motor model on which the present invention relies is expressed as a third-order differential equation. Due to the difficulty of analytical solution, we tend to use numerical solution to explore its dynamic characteristics. The core of this method is to set a fixed step size h , through iterative method, the numerical solution of each time point is calculated successively to deal with this mathematical model which is difficult to directly analyze.

[0099] Among many numerical solution strategies, the Runge-Kutta method is widely used due to its advantages such as clear structure, fast iteration speed and high calculation accuracy. In view of this, the present invention particularly selects the 4th-order Runge-Kutta algorithm to deal with the dynamic problems of induction motors. The specific steps of the method include setting an appropriate step size, and gradually approximating and solving the state of the motor at different time points through a series of iterative calculations. This method is particularly suitable for complex models such as third-order induction motors.

[0100] In S2, both the induction motor sub-model and the static load sub-model are modeled by using a parameter identification method, wherein the induction motor sub-model is modeled by using a parameter identification method, including the following sub-steps:

[0101] S2-1: Calculate the initial voltage, initial active power, initial reactive power, initial power of the induction motor and initial power of the ZIP load model on the load bus, and initialize the intermediate parameters of the induction motor and the ZIP load model;

[0102] S2-2: Based on the initialization results, the fourth-order Runge-Kutta method is used to solve the induction motor differential equation to obtain the induction motor state variables:

[0103] The calculation formula of the induction motor state variable in S2-2 is:

[0104]

[0105] in, and Induction motor in transient process Axis and The electric potential of the axis, is the angular velocity of the rotor, is the number of estimates in the Runge-Kutta method, is the time interval in the Runge-Kutta method, , , , , , , , , , , and represents each partial differential component;

[0106]

[0107]

[0108]

[0109]

[0110] S2-3: Based on the state variables of the induction motor, the power consumption of the induction motor is solved by using the transient potential of the induction motor, and the criterion function is calculated according to the power consumption of the induction motor;

[0111] The criterion function in S2-3 The calculation formula is:

[0112]

[0113] in, is the data sampling length, and are the measured active power and reactive power respectively, and The induction motor sub-model responds to active power and reactive power, respectively.

[0114] S2-4: Use the simulated evolution method to identify the model parameters. If the criterion function is greater than the threshold, return to step S2-2, otherwise complete the parameter identification.

[0115] In the field of power, Bayesian estimation theory is widely used in various scenarios, such as equipment fault detection, maintenance prediction, system reliability assessment, power load forecasting, etc. By utilizing historical data and expert knowledge (prior information), the Bayesian method can provide more accurate estimates of fault probability, optimize maintenance plans, and reduce downtime and costs. In addition, it can handle the uncertainty of complex systems and adapt to changing environmental conditions.

[0116] In the generalized load modeling based on Bayesian estimation theory, the generalized load is divided into four sub-models, namely, the induction motor load model, the static load model, the photovoltaic grid-connected model and the energy storage grid-connected model. The Bayesian estimation theory is used to calculate the probability of the four sub-models appearing in the generalized comprehensive load. The present invention regards this probability as the proportion of a single sub-model load in the generalized comprehensive load. The active power data of the generalized comprehensive load is regarded as a continuous unknown parameter in the Bayesian estimation theory. θ .

[0117] When faced with multi-model dynamic load modeling, the posterior probability of the sub-model is:

[0118]

[0119] in, P x is the output active power of the sub-model, N is the total number of sub-models, s The number of iterations to calculate for the Bayesian estimate, For s -The prior probability obtained after 1 iteration, For sub-model P x exist s After iterations and sub-model P x The residual conditional probability in the matching case, For s After iterations and sub-model P y The residual conditional probability in the matching case, For sub-model P y exist s - Prior probability obtained after 1 iteration.

[0120]

[0121] The above formula is a set vector composed of the residuals of the actual measured value of the generalized comprehensive load active power and the active power output value of each sub-model.

[0122] In the multi-model load modeling method proposed in this invention, it is assumed that In normal distribution, we have:

[0123]

[0124] When the probability of one of the sub-models converges to 1 and the probabilities of the other sub-models converge to 0, if the probability of the model changes, the range of change does not include the sub-model with a probability of 0. Therefore, a critical value is set δ , judge when ,make , to ensure that all future sub-models can be used.

[0125] The Bayesian estimation theory is used to calculate the probability of the sub-model, which is regarded as the proportion of the generalized load model components, taking into account the conditions before and after the time. When it is applied to multi-model load modeling, the probability of the sub-model with higher accuracy at the current moment is calculated based on the probability information of the sub-model at the previous moment and the output error of the sub-model at the current moment. Figure 7 As shown, For real-time voltage data, is the predicted active data of sub-models 1 to N, for The residual matrix associated with the actual active power measurements, ) is in the The predicted active data after iterations, is the measured power data. The overall modeling idea is divided into two parts: the establishment of the sub-model library and the online matching of the sub-model. The establishment of the sub-model is the basic part. First, different types are divided according to different loads (including distributed power sources), and then a set of historical measurement data in the photovoltaic grid-connected system and the energy storage system are selected as the training data of the dynamic RBF network model. The historical measurement data in the motor and constant impedance are selected as the basis for parameter identification, and the trained photovoltaic grid-connected model and the motor, constant impedance, and energy storage system models are used as sub-models. In the online dynamic modeling, the voltage measurement value, the power measurement value, and the voltage change at the current time are used as input information. Each sub-model outputs the power value at the next moment according to the input information table. Then the Bayesian estimation theory calculates the probability of the next moment according to the output error of each sub-model and the probability at the current moment. Finally, the output of all sub-models is integrated in a weighted manner. The integration result is the model output result of the generalized comprehensive load.

[0126] The S3 includes the following sub-steps:

[0127] S3-1: using the voltage measurement value, power measurement value and voltage change at the current moment in each established sub-model as input information of each sub-model, and using each sub-model to obtain the output power value according to the input information;

[0128] S3-2: Calculate the probability of each submodule at the next moment based on the output error of the output power value of each submodel and the probability at the current moment using the Bayesian estimation theory;

[0129] S3-3: Weight the probability of each submodule at the next moment in proportion to obtain the output result of the generalized comprehensive load model of the distribution system.

[0130] The next moment probability of each submodule in S3-2 is:

[0131]

[0132]

[0133] in, For sub-model In the The posterior probability of matching in the iteration, For sub-model In the The prior probability of matching in the iteration, For sub-model In the The prior probability of matching in the iteration, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The output covariance diagonal matrix, For sub-model The output covariance diagonal matrix, is the total number of sub-models, is the critical value set, the superscript is the transpose of the matrix, represents the likelihood function associated with the submodel output error.

[0134] In one embodiment of the present invention, in order to verify the proposed generalized load modeling method based on Bayesian estimation theory, an IEEE 14-node 10kV distribution network simulation test system is built, in which 4 Vf-controlled photovoltaics, 5 induction motors and 4 energy storage batteries are connected respectively, and the connection locations are as follows: Figure 8 As shown, , , and All represent photovoltaic models, All are loads, , , and All are energy storage models. The simulation time is set to 3s, the simulation step is 1µs, the sampling frequency is 1ms, and the photovoltaic model sets the light intensity to be constant at 1000kW / m2 and the temperature to be constant at 25°C. The present invention uses the photovoltaic inverter controlled by Vf as a representative to verify the rationality of the method. If the system contains other controlled inverters, it is only necessary to increase the type of sub-model based on this method.

[0135] exist Figure 8 In the IEEE14-node generalized load access topology diagram, a series of three-phase short-circuit faults of different degrees were implemented at bus node 2, ranging from 15% to 25%. These faults started at 1 second and lasted for 1 second. During this disturbance, focusing on the grid connection of node 1, the per-unit values ​​of active power and reactive power under the influence of the fault were recorded. These data are used as the benchmark measurement values ​​for evaluating the generalized integrated load power of the distribution system under the system disturbance state.

[0136] like Fig. 9 and Fig.10 As shown in the figure, they are respectively the dynamic responses of active power and reactive power of the generalized comprehensive load after the voltage drops by 20%. It can be seen from the figure that compared with the CLM model adopted by the traditional method, the proposed modeling method can better describe the reactive power of the generalized comprehensive load, and the overall change trend of the power curve is very close to the measured value. The output power error under 20% voltage disturbance is shown in Table 1. It can be seen from Table 1 that the active and reactive power simulation errors of the generalized comprehensive load of the present invention are both lower than 5%, and the power simulation errors of the generalized comprehensive load of the traditional CLM model are higher than 5%.

[0137] Table 1 Power error when voltage drops by 20%

[0138]

[0139] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the invention.

Claims

1. A generalized integrated load modeling method for distribution system based on Bayesian estimation theory, characterized in that: The following steps are involved: S1: Obtain the generalized comprehensive load of the distribution system, including the photovoltaic grid-connected sub-model, energy storage grid-connected sub-model, induction motor sub-model and static load sub-model; S2: RBF neural network is used to model the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model, and parameter identification method is used to model the induction motor sub-model and the static load sub-model; S3: Based on the modeling results, the probability of each sub-model appearing in the generalized comprehensive load is calculated using the Bayesian estimation theory, the output result of the generalized comprehensive load model of the distribution system is obtained, and the generalized comprehensive load modeling of the distribution system based on the Bayesian estimation theory is completed; The S3 includes the following sub-steps: S3-1: using the voltage measurement value, power measurement value and voltage change at the current moment in each established sub-model as input information of each sub-model, and using each sub-model to obtain the output power value according to the input information; S3-2: Calculate the probability of each submodule at the next moment based on the output error of the output power value of each submodel and the probability at the current moment using the Bayesian estimation theory; S3-3: weight the probability of each submodule at the next moment in proportion to obtain the output result of the generalized comprehensive load model of the distribution system; The probability of each submodule at the next moment in S3-2 is: in, For sub-model In the The posterior probability of matching in iterations, For sub-model In the The prior probability of matching in the iteration, For sub-model In the The prior probability of matching in the iteration, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The residual matrix associated with the actual active power measurement in, For sub-model The output covariance diagonal matrix, For sub-model The output covariance diagonal matrix, is the total number of sub-models, is the critical value set, the superscript is the transpose of the matrix.

2. The generalized integrated load modeling method for distribution system based on Bayesian estimation theory according to claim 1 is characterized in that: The photovoltaic grid-connected sub-model includes a PV array, a driving Boost circuit, a three-phase voltage source inverter, an LCL filter and a power grid connected in sequence; The energy storage grid-connected sub-model includes an energy storage battery, an inverter circuit, an L-type filter circuit and a power grid connected in sequence; The induction motor sub-model is represented by a third-order induction motor model. The rotor voltage equation of the third-order induction motor model is: in, and They are respectively the motor's transient state Axis and Axis potential, For time, is the transient open circuit time constant, is the rotor open-circuit reactance, is the transient reactance of the induction motor, and The stator winding current is Axis and The weight on the axis, is the synchronous angular velocity of the rotor, is the actual angular velocity of the rotor, and is the stator winding leakage reactance, is the magnetizing reactance, is the rotor winding resistance; The stator current equation of the third-order induction motor model is: in, is the stator winding resistance, and The stator winding terminal voltage is Axis and Components on the axis; The rotor motion equation of the third-order induction motor model is: in, is the motor rotor inertia time constant, is the slip rate of the motor, is the electromagnetic torque of the motor, is the mechanical torque of the motor, and They are the stator current in the transient process. Axis and The weight of the axis, is the initial load rate of the induction motor, and is the mechanical torque characteristic parameter; The static load sub-model is expressed using the ZIP load model as follows: in, and are the active power and reactive power consumed by the ZIP load model, and They are The active power and reactive power consumed by the load model, is the actual operating voltage, is the preset reference voltage, , and They represent the active contribution of the load respectively, , and They represent the reactive contribution of the load respectively.

3. The generalized integrated load modeling method for distribution system based on Bayesian estimation theory according to claim 1 is characterized in that: In S2, the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model are both modeled using RBF neural networks. The input layers of the photovoltaic grid-connected sub-model and the energy storage grid-connected sub-model both receive voltage and frequency change signals from the power grid, and the output layers both output dynamic active power and reactive power signals of the photovoltaic grid-connected power generation system.

4. The generalized integrated load modeling method for distribution system based on Bayesian estimation theory according to claim 1 is characterized in that: In S2, both the induction motor sub-model and the static load sub-model are modeled by using a parameter identification method, wherein the induction motor sub-model is modeled by using a parameter identification method, including the following sub-steps: S2-1: Calculate the initial voltage, initial active power, initial reactive power, initial power of the induction motor and initial power of the ZIP load model on the load bus, and initialize the intermediate parameters of the induction motor and the ZIP load model; S2-2: Based on the initialization results, the fourth-order Runge-Kutta method is used to solve the induction motor differential equation to obtain the induction motor state variables: S2-3: Based on the state variables of the induction motor, the power consumption of the induction motor is solved by using the transient potential of the induction motor, and the criterion function is calculated according to the power consumption of the induction motor; S2-4: Use the simulated evolution method to identify the model parameters. If the criterion function is greater than the threshold, return to step S2-2, otherwise complete the parameter identification.

5. The generalized integrated load modeling method for distribution system based on Bayesian estimation theory according to claim 4 is characterized in that: The calculation formula of the induction motor state variable in S2-2 is: in, and Induction motor in transient process Axis and The electric potential of the axis, is the angular velocity of the rotor, is the number of estimates in the Runge-Kutta method, is the time interval in the Runge-Kutta method, , , , , , , , , , , and represents each partial differential component; 。 6. The generalized integrated load modeling method for distribution system based on Bayesian estimation theory according to claim 5 is characterized in that: The criterion function in S2-3 The calculation formula is: in, is the data sampling length, and are the measured active power and reactive power respectively, and The induction motor sub-model responds to active power and reactive power, respectively.

Images