Active power distribution network dq decoupling model sample expansion information mapping construction method

Abstract

The invention discloses an active power distribution network dq decoupling model sample expansion information mapping construction method, and belongs to the technical field of power distribution networks. The invention aims to provide an active power distribution network dq decoupling model sample expansion information mapping construction method for simulating and expanding limited working condition test sample data of a power distribution area system by using an offline mechanism so as to construct an online physical-information mapping model of the power distribution area system. According to the method, firstly, a multi-element equipment mechanism mapping model is constructed, a dq decoupling mapping model is identified and corrected, and finally, a physical-information data mapping model is constructed based on expansion of a small sample space of the mapping model. The method is separated from a commercialized mechanism simulation system, the modeling cost is saved, the modeling efficiency is improved, and the method is used for online mapping model construction and intelligent analysis decision making of the smart power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution networks. Background Art

[0002] The distribution substation area is developing rapidly towards informatization, intelligence and wisdom. High-fidelity data modeling is an important technical direction. It refers to the establishment and simulation of a physical entity, process or system in an information platform, and the means of analysis by establishing a mapping model of the physical entity in the information space. The core and difficulty of the application of high-fidelity data modeling for power systems lies in the construction of an online high-precision mapping model. The data modeling method is based on a large amount of input and output data, and with the help of appropriate algorithms, it estimates the internal structure and process of the system and establishes a mapping between output and input. Physical-information mapping technology will play an important role in the development of new power systems. It can use virtual and real data fusion and two-way circulation to map the real power system into a virtual digital system driven by data and algorithms, thereby promoting comprehensive perception and intelligent application of the power grid.

[0003] The structure of the distribution substation mainly includes control units composed of a variety of power electronic equipment such as distributed renewable energy generation, distributed energy storage equipment, distributed charging piles, and controllable loads. Various devices are connected through the network to form a power system. For such a system, the study of mechanism or data modeling methods alone has great limitations. The system is too complex and the performance degradation of the equipment after operation generally leads to the limited modeling accuracy of the simple mechanism modeling method, and it relies on the commercial software system of mechanism simulation; while the simple data modeling method has high cost and long cycle to obtain multi-condition test data, and there may be model failure problems when the extended condition is applied. Therefore, the modeling method that combines mechanism and data methods has become an important way to study complex distribution substations. Summary of the invention

[0004] The purpose of the present invention is to propose a method for constructing an active distribution network dq decoupling model sample expansion information mapping by using offline mechanism simulation to expand the finite operating condition test sample data of the distribution substation system, and then constructing an online physical-information mapping model of the distribution substation system.

[0005] The steps of the present invention are: S1. Construction of multi-device mechanism mapping model based on dq0 decoupling model: The dq0 transform maps the sinusoidal signal to a constant, thus producing a relatively simple dynamic model. The dq0 transform and its inverse function are defined as follows: Where, angle θ is the reference angle or reference phase; S11, overall system state modeling based on dq0 coordinate system The overall state of the system is represented by a linear state space model of voltage input and current output: Where ξ is the state vector; The input and output vectors are: In the formula, v d,n ,v q,n ,v 0,n is the dq0 voltage of bus n, i d,n ,i q,n ,i 0,n is the dq0 current injected into bus n from the nth unit; S12, load modeling based on the dq0 coordinate system The three-phase resistance based on symmetrical configuration in the dq0 coordinate system is: The three-phase inductance based on symmetrical configuration in the dq0 coordinate system is: The three-phase capacitors with symmetrical configuration based on the dq0 coordinate system are: in: Where, angle θ is the reference angle or reference phase; S13, the dynamic equation of linear circuit modeling based on the dq0 coordinate system is: S14. The basic dynamic model of distributed energy storage modeling based on the dq0 coordinate system is: Where E(t) is the stored energy; P(t) is the total power flowing into the device; η c (E,P) is the charging efficiency; η d (E,P) is the discharge efficiency; The ratio of the energy extracted from the device to the energy stored in the device, i.e. the round-trip efficiency η r , if η c and η d is a constant, then: The current of the energy storage system is decomposed into d-axis and q-axis components, as well as a zero-sequence component: i ES =i ES,d +i ES,q +i ES,0 (13) The SOC balance of the energy storage system is expressed by equation (14): Where SOC(t) is the state of charge of the energy storage system at time t, P ES,c (t) and P ES,d (t) are the charging and discharging power of the energy storage device at time t, and C is the capacity of the energy storage system; S15. Distributed photovoltaic modeling based on dq0 coordinate system The maximum power point tracking control strategy is adopted in the dq0 coordinate system: P mppt =S·λ PV (15) Where P mppt is the power of the photovoltaic array in MPPT mode; S is the light intensity; λ PV is the light intensity coefficient; Photovoltaic cell current Where U pv is the photovoltaic array voltage in mppt mode; In the dq0 coordinate system, the control strategy of the inverter is: In the formula, k ii is the gain of the integral controller; k ip is the gain of the proportional controller; L is the equivalent inductance of the inverter; S2, identification and correction of the dq decoupling mapping model based on data By using limited operating data, the input parameters of the distribution station area status are imported into the model, and the model is simulated using a mechanism simulation system. The output characteristic parameters of the operation calculation are compared with the characteristic parameters in the operating data, and the model is identified and corrected. S3. Expansion of small sample space based on mapping model The sample mean expectation and variance representing the operating state of the power grid under specific conditions are as follows: Assume X is a random variable representing a state variable of the power grid, X1,X2,...,X N are N independent and identically distributed samples of X, with sample mean Defined as: When the sample size N is large enough, the sample mean Will approach the true expected value E(X) of X: The output of the solar photovoltaic panel P pv It is expressed as: In the formula, μ pv is the average value of the photovoltaic panel output, is the variance of the PV panel output; Random sampling is performed in Monte Carlo simulation, which is expressed as: X i ~F -1 (U i ) (twenty one) In the formula, X i is the i-th sample value, F -1 is the inverse function of the cumulative distribution function of the state variable, U i is a random number uniformly distributed in the interval [0,1]; Expected value of power grid operation E[Y]: Where N is the number of samples, Y i is the grid operation status obtained from the ith simulation; The sample space of the unit model before and after sample space expansion is as follows: Ω={θ1,θ2…θ i} (twenty three) Ω'={θ1,θ2…θ n} (twenty four) In the formula, θ is the data sample obtained by simulating the mechanism model under a certain working condition; Ω and Ω' represent the sample space of the mechanism model before and after expansion; the subscript is the number of the working condition, where i is much smaller than n; S4. Construction of physical-information data mapping model The structural layers of the neural network are connected by weight multiplication and bias. The nodes of two adjacent layers are connected by full connection, that is, each node in the previous layer is connected to each node in the next layer through weight w and bias. The relationship is expressed as: y=f(ω,x) (25) In the formula, y is the output matrix; x is the input matrix; ω is the weight matrix; f is the nonlinear activation function; When the distribution area system uses the BP neural network algorithm to build a data model, its input and output matrices are expressed as: B=(P pv ,Q charge ,Q discharge ,...) (26) Y=(S storage ,Q grid ,P total,…) (27) Where P pv is the matrix of solar photovoltaic output, in kW; Q charge ,Q discharge are matrices composed of the charging and discharging flows of energy storage devices, in kW; S storage is the matrix of battery energy storage state; Q grid P is the matrix of power supply quality parameters in the distribution area; total is the total power generation matrix of the power grid; After collecting multiple sets of derived operating condition data and performing necessary preprocessing on them, the data are organized into data sets, and the relationship mapping between input and output is established using data modeling methods.

[0006] The present invention is independent of commercial mechanism simulation systems, saves modeling costs, improves modeling efficiency, and is used for online mapping model construction and intelligent analysis and decision-making of smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is the flow chart of modeling the distribution substation area system; Figure 2 Signal flow diagram of the overall status of the transmission network; Figure 3 Schematic diagram of load model based on dq0 coordinate system Figure 4 Schematic diagram of linear circuit based on dq0 coordinate system; Figure 5 Schematic diagram of energy storage equipment; Figure 6 Schematic diagram of photovoltaic equipment; Figure 7 The structure of a neural network; Figure 8 The construction process of the physical-information mapping model of the distribution area system based on the BP neural network method; Fig. 9 Wiring diagram of power system mechanism model based on dq0 model; Fig.10 Operation effect diagram of power system mechanism model based on dq0 model; Fig.11 The error of the physical-information mapping model relative to the mechanism mapping model; Fig.12 The error of the physical-information mapping model under different test sets; Fig.13 The error between the physical information mapping model and the actual working conditions. DETAILED DESCRIPTION

[0008] In view of the high technical cost of the online high-precision mechanism simulation system, the high cost and time-consuming problem of the multi-operating test data modeling method, this invention will propose a method of using offline mechanism simulation to expand the limited operating condition test sample data of the distribution substation system, and then construct an online physical-information mapping model of the distribution substation system, so as to break away from the commercial mechanism simulation system, save modeling costs, improve modeling efficiency, and be used for online mapping model construction and intelligent analysis and decision-making of smart grids. The purpose is to map the physical models of renewable energy power generation equipment, energy storage equipment, controllable loads, and control units composed of various power electronic devices in the distribution substation to digital information models through mapping technology.

[0009] The steps of the present invention are: (1) Construction of the mechanism model based on the dq0 decoupling model: The distribution network system transmits information and energy through power, communication signals, etc., so as to connect and couple the scattered devices to form an overall distribution network system. The dq0 model-based method uses the known system structure and dynamic equations to create a simplified model that still captures the main characteristics of the system while retaining the physical meaning and structure of the system. The dq0 model converts the three-phase alternating current (AC) signal in the three-phase power system into two DC quantities (d axis and q axis) and a zero-sequence component (0 axis) through mathematical transformation. This conversion is based on mathematical abstractions such as Clarke transformation, which enables the complex three-phase AC system to be analyzed and controlled in a more simplified two-dimensional space, so that it maintains accuracy at high frequencies while avoiding the assumptions and limitations of quasi-static networks. It can handle nonlinear systems, thereby performing dynamic simulation and stability analysis. Based on this, the construction of a mechanism model based on the dq0 model can effectively reflect the state of each component within the distribution station area and the energy flow relationship within the system.

[0010] (2) Identification and correction of the dq decoupling mapping model based on data: The mechanism modeling at the general system level will focus on the description of the coupling relationship between different modules in the system, while the single device module is simplified into a black box structure that only considers the relationship between the module output and input. However, in addition to the connection structure between the various devices, the internal structure and operating status of the individual devices in the actual power system are also extremely complex. Moreover, with its long-term operation, the status of the equipment is in a state of performance degradation and change. Therefore, by using limited operating data (several or more than a dozen groups), the unit state input parameters are imported into the model, and the mechanism simulation system is used to simulate the operation, the output characteristic parameters of the operation calculation are compared with the characteristic parameters in the operation data, and the model is identified and corrected to achieve the construction of a mechanism mapping model with high accuracy. The operating data used for correction is generally obtained through field tests, or directly selecting representative operating conditions, and directly obtained through the DCS (Distributed Control System) data monitoring and acquisition system of various parts of the system.

[0011] (3) Expansion of small sample space based on dq mapping model: The expansion method of the sample space of the mechanism mapping model is to simulate multiple working conditions or even all working conditions of the mechanism mapping model based on the mechanism simulation system to obtain the multi-working condition sample space of the distribution network system. The sample of a single working condition should include parameters such as voltage, current, power, etc. of all nodes in the distribution network system, as well as the operating characteristics of various types of equipment, such as the transformation ratio and load of the transformer, and the status of the switchgear. Through multiple simulation operations of the mechanism simulation system, the operating data of only a few groups of working conditions can be expanded to sample spaces of multiple working conditions or even the full working condition operation range, thereby meeting the data volume requirements for system data modeling.

[0012] (4) Construction of physical-information data mapping model: The key to data modeling is the process of training the model, which requires the use of appropriate algorithms. In the distribution network system, there are very complex nonlinear relationships between various parameters. Traditional data models cannot effectively extract the nonlinear relationship between input and output. The machine BP (Back Propagation) neural network has strong self-organization, self-learning, self-adaptation and nonlinear mapping capabilities, so it can be used as a suitable choice for constructing the physical-information mapping model of the distribution system. In the distribution network system, load fluctuations, changes in power quality and the access of distributed energy are the main reasons for changes in the system operation status. Therefore, in the process of constructing the mapping model, the load, various power outputs and related network parameters (such as voltage, current, etc.) should be selected as the input matrix; while the power supply reliability, equipment operation status and frequency changes of the distribution network are important characteristic parameters that reflect the operation status of the distribution network system and should be used as components of the output matrix.

[0013] (5) Verification of the accuracy of the data mapping model: After collecting multiple sets of derived operating condition data and performing necessary preprocessing on them, the data are organized into data sets, and the relationship mapping between input and output is established using data modeling methods. Based on the constructed physical-information mapping model, the existing limited operating conditions of the unit are simulated, and the error calculation results of the simulation output (physical-information mapping) are compared with the operating condition parameters (mechanism mapping) to prove their accuracy.

[0014] The main feature of the present invention is to propose a method for constructing a physical-information mapping model based on a dq0 decoupling model expansion sample, build a mechanism model based on the dq0 model, and establish a basic framework model of a distribution station area; use the limited working condition operation data identification and correction to obtain a corrected high-precision mechanism mapping model; on the basis of the high-precision mechanism mapping model, expand the limited operation data sample through the mechanism simulation system to obtain operation data of multiple working conditions or even the entire working condition range; finally determine the input state variables and output characteristic parameters of the unit, and use the data modeling method to construct a physical-information mapping model between input and output to replace the mechanism mapping model that relies on a high-cost mechanism simulation system. The proposed technology can effectively deal with the system complexity is too large and the performance degradation of the equipment after operation generally leads to the limited modeling accuracy of the simple mechanism modeling method, and the commercial software system relying on mechanism simulation and the simple data modeling method have high cost and long cycle to obtain multi-condition test data, and there may be model failure problems when the extended working condition is applied.

[0015] The present invention is described in further detail below: (1) Construction of multi-device mechanism mapping model based on dq0 decoupling model The power grid system transmits matter, information and energy through media such as electricity, information flow and control signals, so as to connect and couple the scattered devices to form an overall distribution network system. The fundamental content of the power grid system mechanism model is the simulation of the state and parameter changes of electricity in the process of flowing between devices and networks. The distributed multi-device mechanism mapping model under the dq0 coordinate axis describes the power system through a signal flow graph, in which each component is modeled by a dq0 quantity.

[0016] The dq0 transform maps a sinusoidal signal to a constant, thus producing a relatively simple dynamic model. The dq0 transform and its inverse function are defined as follows: Wherein, angle θ is the reference angle or reference phase.

[0017] 1) Modeling the overall state of the system based on the dq0 coordinate system The overall state signal flow diagram of the transmission network is as follows: Figure 2 The overall state of the system is represented by a linear state space model of voltage input and current output: Where ξ is the state vector.

[0018] The input and output vectors are: In the formula, v d,n ,v q,n ,v 0,nis the dq0 voltage of bus n, i d,n ,i q,n ,i 0,n is the dq0 current injected into bus n from the nth unit.

[0019] 2) Load modeling based on dq0 coordinate system In a three-phase circuit, a balanced three-phase signal is a sinusoidal signal with equal amplitude and phase shift of ±120°. The load modeling diagram based on the dq0 coordinate system is as follows: Figure 3 As shown. The three-phase resistance based on symmetrical configuration in the dq0 coordinate system is:

[0020] The three-phase inductance based on symmetrical configuration in the dq0 coordinate system is:

[0021] The three-phase capacitors with symmetrical configuration based on the dq0 coordinate system are: in: Wherein, angle θ is the reference angle or reference phase.

[0022] 3) Linear circuit modeling based on dq0 coordinate system Taking the symmetrical configuration RL transmission line as an example, the circuit diagram is as follows Figure 4 Its dynamic equation is: The model obtained based on the dq0 coordinate system is linear and time-invariant.

[0023] 4) Distributed energy storage modeling based on dq0 coordinate system Distributed energy storage devices based on the dq0 coordinate system, such as Figure 5 As shown. The basic dynamic model of the device is: Where E(t) is the stored energy; P(t) is the total power flowing into the device; η c (E,P) is the charging efficiency; η d (E,P) is the discharge efficiency.

[0024] The energy derivative (absolute value) is higher than the power flowing into the grid. In addition, the round-trip efficiency (η r ) is the efficiency of a charge-discharge cycle. Consider a device that low Charge to E high , then from E high Discharge to E lowThe round-trip efficiency is the ratio of the energy extracted from the device to the energy stored in the device. If η c and η d is a constant, then:

[0025] In the dq0 coordinate system, the current of the energy storage system can be decomposed into d-axis and q-axis components, as well as the zero-sequence component: i ES =i ES,d +i ES,q +i ES,0 (13)

[0026] The SOC (State of Charge) balance of the energy storage system can be expressed by equation (14): Where SOC(t) is the state of charge of the energy storage system at time t, P ES,c (t) and P ES,d (t) are the charging and discharging power of the energy storage device at time t, and C is the capacity of the energy storage system.

[0027] 5) Distributed photovoltaic modeling based on dq0 coordinate system Photovoltaic array model: Modeling photovoltaic devices in the dq0 coordinate system usually refers to modeling the control of photovoltaic arrays and inverters. The photovoltaic array can be modeled as a current source whose output current depends on the light intensity and temperature, such as Figure 6 shown.

[0028] The MPPT (maximum power point tracking) control strategy is adopted in the dq0 coordinate system P mppt =S·λ PV (15) Where P mppt is the power of the photovoltaic array in MPPT mode; S is the light intensity; λ PV is the light intensity coefficient.

[0029] Photovoltaic cell current as follows: Where U pv is the photovoltaic array voltage in mppt mode.

[0030] Inverter model: The inverter is usually modeled as a voltage source whose output voltage can be controlled to match the amplitude and phase of the grid voltage. In the dq0 coordinate system, the control strategy of the inverter can be expressed as: In the formula, k ii is the gain of the integral controller; k ip is the gain of the proportional controller; L is the equivalent inductance of the inverter.

[0031] (2) Identification and correction of the dq decoupling mapping model based on data The mechanism mapping model established based on the distribution substation system structure and the basic equations of the power flow process also needs to be identified and corrected using limited operating condition data to build a mechanism mapping model with high accuracy. The general system-level mechanism modeling focuses on the description of the coupling relationship between different modules in the system, while the single device module is simplified to a black box structure that only considers the relationship between the module output and input. However, in addition to the connection structure between each device, the actual power grid system also has extremely complex internal structures and operating states of a single device. Moreover, with the long-term operation of the system, the state of the equipment is in performance degradation and change. Therefore, the accuracy of the mechanism mapping model established only based on the power grid system structure and basic equations is questionable.

[0032] The present invention uses limited operating data (several or more than ten groups), imports the input parameters of the distribution station area status into the model, uses the mechanism simulation system to simulate the operation, compares the output characteristic parameters of the operation calculation with the characteristic parameters in the operating data, and identifies and corrects the model to achieve the construction of a mechanism mapping model with high accuracy. The operating data used for correction is generally obtained directly through field tests or by directly selecting representative operating conditions and through the monitoring system of the power grid.

[0033] (3) Expansion of small sample space based on mapping model After building an accurate grid mechanism mapping model, we are faced with the problem of how to expand the sample space to meet the needs of data modeling. Traditional data modeling methods rely on a large amount of operating data, and the limited data we have is obviously not enough to support this demand. Therefore, we plan to use Monte Carlo simulation technology to expand our sample library.

[0034] Monte Carlo simulation is a statistical method that uses random sampling to approximate the solution of deterministic problems. The theoretical basis of the Monte Carlo method is the law of large numbers, which describes the results of a considerable number of repeated experiments. According to this law, the larger the number of samples, the closer the average is to the true value. Monte Carlo is used in the power grid model to simulate the operation of the power grid under various conditions to generate more usable sample data.

[0035] First, define the state variables of the power grid, such as the output of solar photovoltaic panels and the charging and discharging status of energy storage devices. Then build probability distribution models for these variables, which will be based on the physical characteristics of the power grid and historical data. Based on this, use the Monte Carlo method to generate a large number of random samples.

[0036] Each sample represents the operating state of the power grid under specific conditions. The expectation and variance of the sample mean are as follows: Let X be a random variable representing a state variable of the power grid, X1, X2, ..., X N are N independent and identically distributed samples of X. The sample mean is Defined as:

[0037] According to the law of large numbers, when the sample size N is large enough, the sample mean Will approach the true expected value E(X) of X:

[0038] In the power system, we first need to define the state variables that affect the operation of the system, such as the output P of the solar photovoltaic panel. pv The probability distribution model of these variables can be established based on the physical characteristics of the power grid and historical data. It can be expressed as: In the formula, μ pv is the average value of the photovoltaic panel output, is the variance of the PV panel output.

[0039] Random sampling is performed in Monte Carlo simulation, which can be expressed as: X i ~F -1 (U i )(twenty one) Where, X i is the i-th sample value, F -1 is the inverse function of the cumulative distribution function of the state variable, U i is a random number uniformly distributed in the interval [0,1].

[0040] The distribution characteristics of statistical output variables are such as calculating the expected value E[Y] of the power grid operation: Where N is the number of samples, Y i is the grid operation status obtained by the ith simulation.

[0041] Based on this, the limited operating condition data can be expanded to a wider range of operating conditions, thereby meeting the needs of data modeling and improving the predictive ability and generalization of the power grid model. The sample space of the unit model before and after the sample space expansion is as follows: Ω={θ1,θ2…θ i} (twenty three) Ω'={θ1,θ2…θ n} (twenty four) Where θ is the data sample obtained by simulating the mechanism model under a certain working condition; Ω and Ω' represent the sample space of the mechanism model before and after expansion; the subscript is the number of the working condition, where i is much smaller than n.

[0042] (4) Construction of physical-information data mapping model The rapid development of computer technology has made it possible for people to conduct centralized analysis of large amounts of data, and data modeling has become a common analytical method in the industrial field. The key to data modeling is the process of training the model, which requires the use of appropriate algorithms. There are very complex nonlinear relationships between various parameters in the power system. Traditional data models such as linear regression cannot effectively extract the nonlinear relationship between input and output. BP (Back Propagation) neural networks have strong self-organization, self-learning, self-adaptation and nonlinear mapping capabilities, so they can be used as a suitable choice for building a distribution substation system mapping model. The structure of the neural network mainly includes an input layer, a hidden layer and an output layer. The basic unit of each layer is a neuron, and the nodes of adjacent layers are fully connected. The input layer receives the input multidimensional data X, the hidden layer is generally a single layer or multi-layer neuron to extract and transform the input data, and the output layer outputs the final result Y. Generally speaking, the number of nodes in the input layer and the output layer is fixed, and the number of hidden layers can be freely specified.

[0043] Layers are connected by weight multiplication and bias, and nodes in two adjacent layers are connected by full connection, that is, each node in the previous layer is connected to each node in the next layer through weight w and bias. The relationship can be expressed as: y=f(ω,x) (25) In the formula, y is the output matrix; x is the input matrix; ω is the weight matrix; f is the nonlinear activation function.

[0044] The modeling process of BP neural network: first initialize the weights and biases of each node; secondly input the training data and calculate the output value of each node through forward propagation; then calculate the deviation function through the output value and the true value, and then calculate the negative gradient of each parameter through back propagation based on the predicted deviation; finally update the gradient and bias. Iterate the above steps until the error converges, and the BP neural network model training is completed.

[0045] In the power grid system, the fluctuation of power load, the change of renewable energy output and the charging and discharging state of energy storage equipment are the main reasons for the change of system working conditions. Therefore, in the process of constructing the data mapping model, the output of distributed power generation equipment, the charging and discharging amount of energy storage equipment and the related power flow (such as battery charging and discharging rate) in each mechanism model sample should be selected to form the input matrix; while the battery energy storage state, power supply quality parameters and total power generation are important characteristic parameters reflecting the operating state of the microgrid system, and should be used as the content of the output matrix.

[0046] When the distribution area system uses the BP neural network algorithm to build a data model, its input and output matrices can be expressed as: B=(P pv ,Q charge ,Q discharge ,…) (26) Y=(S storage ,Q grid ,P total ,…) (27) Where P pv is the matrix of solar photovoltaic output, in kW; Q charge ,Q discharge are matrices composed of the charging and discharging flows of energy storage devices, in kW; S storage is the matrix of battery energy storage state; Q grid P is the matrix of power supply quality parameters in the distribution area; total is the total power generation matrix of the power grid.

[0047] After collecting multiple sets of derived working condition data and performing necessary preprocessing on them, the data are organized into data sets, and the relationship mapping between input and output is established using data modeling methods. The distribution substation system is a typical multi-input and multi-output system, so this paper uses the BP neural network algorithm. The process of establishing the data model can be as follows: Figure 7 As shown in the figure. In this structure, the input layer receives the input data of the working condition; the hidden layer has multiple layers, each containing multiple neurons; the output layer has multiple fully connected layers in parallel, and each layer outputs multiple parameters to be predicted. Then, according to the actual situation of the data set, the parameters required by the algorithm itself are set, such as the activation function, learning rate, and maximum number of iterations of the hidden layer. Through the above process, we can build a physical-information mapping model of the power grid system. Based on this model, the operation and maintenance personnel of the power grid can calculate the characteristic parameters of the system by inputting the specified system variables without being separated from the mechanism simulation system, and predict and analyze the operating status of the system under different working conditions.

[0048] Simulation Verification (1) Establishment of distribution station area mechanism mapping model based on dq0 decoupling model The present invention constructs a mechanism mapping model of the distribution substation area based on the dq0 decoupling model through power simulation software. Fig. 9 .

[0049] Fig.10 This is the operation effect of the power system mechanism model based on the dq0 model. The system frequency is 50 Hz, and the figure shows the frequency values ​​at bus 4 and bus 5. The active power at bus 3, bus 4 and bus 5 is also shown in the figure. This paper conducts simulation calculations for three typical operating conditions in the distribution substation area, and selects key parameters in the distribution substation area, including distributed power output, battery energy storage status, power supply quality parameters and total power generation, as evaluation indicators. The mechanism mapping model is verified by comparing the errors between the simulation results and the actual parameters of these parameters under different operating conditions. The three groups of operating conditions used for verification include: normal operating conditions at full load and two groups of conditions with different energy distribution at full load. The specific comparison results are shown in Table 1 (G in the table represents the unit load rate, expressed in percentage; PV and Bat represent the distributed power supply and battery energy storage status, respectively, in kilowatts).

[0050] Table 1

[0051] The comparison results show that under all working conditions, the errors between the main parameter results obtained by calculation and the actual working condition parameters are within 3%, which meets the accuracy requirements of the simulation model, thus confirming that the constructed mechanism mapping model has high accuracy and does not require further identification and correction.

[0052] Based on the established high-precision mechanism mapping model, this paper inputs multiple sets of unknown operating conditions and simulates these conditions to expand the sample space of the model. Specifically, this paper simulates the operating conditions of different distributed power sources and battery energy storage states. The extended operating condition sample range covers distributed power output 50-150kW, battery energy storage output 30-120kW, and 1kW as the output change gradient. 100 sets of unknown different energy output conditions are simulated.

[0053] (2) Physical-information mapping model based on multiple operating parameters Based on the expanded model condition samples, the BP neural network method is used to construct a physical-information mapping model. This paper selects the output of distributed generation equipment and the battery energy storage status as the input of the data mapping model, and the total power generation of the distribution station area and the power supply quality parameters of the power grid (such as voltage stability and frequency stability) as the output. After the steps of the aforementioned BP neural network algorithm, a data proxy mapping model is constructed.

[0054] (3) Consistency verification of the physical-information mapping model After the data proxy mapping model is successfully established, more than 100 groups of extended samples in the sample space expanded by the mechanism simulation are used as test sets to test the consistency between the data proxy mapping model and the mechanism mapping model. The results of the three characteristic parameters of distributed power output, total power generation and power supply quality parameters in the mechanism mapping model and the data proxy mapping model are compared, such as Fig.11 As shown in Figure 2, the calculation results of the three characteristic parameters in the two models have very small errors, and the maximum error does not exceed 1%.

[0055] In the same data set, different combinations of test sets and training sets generated by different partitions of the data set may affect the simulation results. In order to avoid interference from different combinations of data sets and further illustrate the consistency between the data proxy mapping model and the mechanism mapping model, this paper conducted four repeated tests under different and random test sets while ensuring the consistency of the sample ratio. The following takes the output of distributed power sources as an example, and the results are shown as follows: Fig.12 shown.

[0056] The errors between the results obtained by each data proxy mapping model and the calculation results of the mechanism mapping model are still very small, with the maximum error less than 1%. Therefore, the constructed data proxy mapping model has a high consistency with the mechanism mapping model and can be used to replace the mechanism mapping model and mechanism simulation system.

[0057] (4) Verification of the accuracy of the physical-information mapping model The simulation results of the physical-information mapping model are highly consistent with those of the mechanism mapping model. To further verify the accuracy of the model, the existing limited operating conditions of the unit are simulated based on the constructed physical-information mapping model, and the simulation output is compared with the operating condition parameters. The results are as follows: Fig.13 As shown. The error between the simulation effect of the physical-information model and the operating parameters is less than 3%, which meets the accuracy requirements of the model. Therefore, the physical-information mapping model established by the present invention has good accuracy, and its simulation results can provide important reference for the design and planning of the unit.

Claims

1. A method for constructing sample extended information mapping of a dq decoupling model of an active distribution network, characterized in that: The steps are: S1. Construction of multi-device mechanism mapping model based on dq0 decoupling model: The dq0 transform maps the sinusoidal signal to a constant, thus producing a relatively simple dynamic model. The dq0 transform and its inverse function are defined as follows: Where, angle θ is the reference angle or reference phase; S11, overall system state modeling based on dq0 coordinate system The overall state of the system is represented by a linear state space model of voltage input and current output: Where ξ is the state vector; The input and output vectors are: In the formula, v d,n ,v q,n ,v 0,n is the dq0 voltage of bus n, i d,n ,i q,n ,i 0,n is the dq0 current injected into bus n from the nth unit; S12, load modeling based on the dq0 coordinate system The three-phase resistance based on symmetrical configuration in the dq0 coordinate system is: The three-phase inductance based on symmetrical configuration in the dq0 coordinate system is: The three-phase capacitors with symmetrical configuration based on the dq0 coordinate system are: in: Where, angle θ is the reference angle or reference phase; S13, the dynamic equation of linear circuit modeling based on the dq0 coordinate system is: S14. The basic dynamic model of distributed energy storage modeling based on the dq0 coordinate system is: Where E(t) is the stored energy; P(t) is the total power flowing into the device; η c (E,P) is the charging efficiency; η d (E,P) is the discharge efficiency; The ratio of the energy extracted from the device to the energy stored in the device, i.e. the round-trip efficiency η r , if η c and η d is a constant, then: The current of the energy storage system is decomposed into d-axis and q-axis components, as well as a zero-sequence component: i ES =i ES,d +i ES,q +i ES,0 (13) The SOC balance of the energy storage system is expressed by equation (14): Where SOC(t) is the state of charge of the energy storage system at time t, P ES,c (t) and P ES,d (t) are the charging and discharging power of the energy storage device at time t, and C is the capacity of the energy storage system; S15. Distributed photovoltaic modeling based on dq0 coordinate system The maximum power point tracking control strategy is adopted in the dq0 coordinate system: P mppt =S·λ PV (15) Where P mppt is the power of the photovoltaic array in MPPT mode; S is the light intensity; λ PV is the light intensity coefficient; Photovoltaic cell current Where U pv is the photovoltaic array voltage in mppt mode; In the dq0 coordinate system, the control strategy of the inverter is: In the formula, k ii is the gain of the integral controller; k ip is the gain of the proportional controller; L is the equivalent inductance of the inverter; S2. Identification and correction of the dq decoupling mapping model based on data By using limited operating data, the input parameters of the distribution station area status are imported into the model, and the model is simulated using a mechanism simulation system. The output characteristic parameters of the operation calculation are compared with the characteristic parameters in the operating data, and the model is identified and corrected. S3. Expansion of small sample space based on mapping model The sample mean expectation and variance representing the operating state of the power grid under specific conditions are as follows: Assume X is a random variable representing a state variable of the power grid, X1,X2,...,X N are N independent and identically distributed samples of X, with sample mean Defined as: When the sample size N is large enough, the sample mean Will approach the true expected value E(X) of X: The output of the solar photovoltaic panel P pv It is expressed as: In the formula, μ pv is the average value of the photovoltaic panel output, is the variance of the PV panel output; Random sampling is performed in Monte Carlo simulation, which is expressed as: X i ~F -1 (U i ) (21) Where, X i is the i-th sample value, F -1 is the inverse function of the cumulative distribution function of the state variable, U i is a random number uniformly distributed in the interval [0,1]; Expected value of power grid operation E[Y]: Where N is the number of samples, Y i is the grid operation status obtained from the ith simulation; The sample space of the unit model before and after sample space expansion is as follows: Ω={θ1,θ2…θ i } (23) Ω'={θ1,θ2…θ n } (24) In the formula, θ is the data sample obtained by simulating the mechanism model under a certain working condition; Ω and Ω' represent the sample space of the mechanism model before and after expansion; the subscript is the number of the working condition, where i is much smaller than n; S4. Construction of physical-information data mapping model The structural layers of the neural network are connected by weight multiplication and bias. The nodes of two adjacent layers are connected by full connection, that is, each node in the previous layer is connected to each node in the next layer through weight w and bias. The relationship is expressed as: y=f(ω,x) (25) In the formula, y is the output matrix; x is the input matrix; ω is the weight matrix; f is the nonlinear activation function; When the distribution area system uses the BP neural network algorithm to build a data model, its input and output matrices are expressed as: B=(P pv ,Q charge ,Q discharge ,…) (26) Y=(S storage ,Q grid ,P total ,…) (27) Where P pv is the matrix of solar photovoltaic output, in kW; Q charge ,Q discharge are matrices composed of the charging and discharging flows of energy storage devices, in kW; S storage is the matrix of battery energy storage state; Q grid P is the matrix of power supply quality parameters in the distribution area; total is the total power generation matrix of the power grid; After collecting multiple sets of derived operating condition data and performing necessary preprocessing on them, the data are organized into data sets, and the relationship mapping between input and output is established using data modeling methods.

Images