Dynamic state estimation method and system for neural network of power distribution network based on digital twinning

Abstract

The invention discloses a power distribution network neural network dynamic state estimation method and system based on digital twinning, and relates to the technical field of power system state estimation, and the method comprises the steps: obtaining real-time sampling data from an actual power distribution network, and uniformly transmitting the data to a control server; a digital twinborn model operated by the control server screens and processes the data sampled in real time, and various data required by dynamic state estimation of the power distribution network are obtained through the digital twinborn model; performing dynamic state estimation on the power distribution network through a recurrent neural network trained in advance according to various data; according to the method provided by the invention, the problems of data transmission and acquisition interference and missing are solved, and required accurate data can be screened and complemented to serve as basic data of dynamic state estimation; compared with traditional Kalman filtering, the method has higher calculation speed and robustness, different prediction models can be adaptively replaced according to different topological structures, and a good prediction effect can be achieved for dynamic state estimation in new energy scenes such as photovoltaic and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of power system dynamic state estimation, and specifically to a distribution network neural network dynamic state estimation method and system based on digital twins. Background Art

[0002] With the widespread integration of distributed renewable energy sources such as wind and solar power into the power grid, developing a low-carbon, sustainable energy system centered on high renewable energy penetration has become a key strategic goal for countries around the world. The influx of rapidly changing, high-power generating equipment and stations has exacerbated the persistent issues of data interference and loss in data transmission within distribution networks. Accurate data is essential for controlling these rapidly changing, yet difficult to coordinate and control, renewable energy generation equipment.

[0003] Meanwhile, traditional physics-based dynamic state estimation methods, such as Kalman filtering, used to mitigate operational risks in distribution networks rely on an accurate dynamic model of the entire system to estimate the system state. However, in distribution networks with a high proportion of renewable energy, a complete physical model is often unavailable due to factors such as the unidentifiable parameters of distributed inverter controllers, frequent changes in distributed power generation control methods and plug-and-play, and data privacy requirements. This complexity leads to uncertain dynamic models of subsystems in distribution networks with a high proportion of renewable energy, which inevitably renders the dynamic state estimation of traditional physics-based Kalman filtering inaccurate or even uncertain. This has a significant negative impact on risk control in distribution networks with a high proportion of renewable energy.

[0004] Therefore, it is necessary to develop new methods that can accurately estimate the dynamic state of distribution networks under different operating conditions based on the screening and correction of large amounts of data, so as to effectively avoid distribution network operation risks. Summary of the Invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] A first aspect of an embodiment of the present invention provides a distribution network neural network dynamic state estimation method based on digital twins, including: obtaining real-time sampled data from an actual distribution network and sending it uniformly to a control server; a digital twin model run by the control server filters and processes the real-time sampled data, and obtains various types of data required for dynamic state estimation of the distribution network through the digital twin model; and performing dynamic state estimation of the distribution network through a recurrent neural network trained in advance based on the various types of data.

[0008] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, it utilizes each acquisition node in the actual distribution network to collect data in a pre-set acquisition cycle and transmits the data collected by each node to the control server in a communication mode of power carrier and wireless network.

[0009] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, wherein: screening and processing the real-time sampled data includes:

[0010] According to the line impedance or admittance in the real-time sampling data obtained from the actual distribution network, the original node admittance matrix Y of the distribution network is formed, and the maximum number of iterations k is set;

[0011] Select the balance node voltage and assign the voltage initial value V0=1 to other PI nodes in the whole network, and set the phase angle to 0. For PV type nodes, set the reactive initial value to 0; for PQ(V) type nodes, set the reactive initial value to Q0=f(V0); for PI type nodes, set the reactive initial value to Where I represents the current injected into the PI node, and P represents the active power injected into the PI node;

[0012] The Newton-Raphson method is used to solve the node voltage and obtain the new iterative node voltage amplitude and phase angle value. For PV type nodes, PQ (V) type nodes and PI type nodes, corrections are made according to their respective reactive power correction equations.

[0013] Verify whether the convergence condition is met according to the correction result;

[0014] The convergence conditions for PQ nodes, PQ(V) nodes, and PI nodes are:

[0015]

[0016] The convergence conditions for PV type nodes are:

[0017]

[0018] in, represents the voltage amplitude of node i at the k+1th iteration, represents the voltage amplitude of node i at the kth iteration, ε represents the preset threshold, V s Indicates the node voltage amplitude of the PV node;

[0019] If the convergence condition is met, the calculation ends and the number of iterations, convergence accuracy, voltage amplitude and phase are output; otherwise, the calculation jumps to the step of solving the voltage of each node and continues.

[0020] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, the acquisition of the node voltage amplitude and phase angle value of the new iteration includes solving the node voltage and calculating the active power and reactive power using the Newton-Raphson method, which is expressed as:

[0021]

[0022]

[0023] Among them, P k represents the active power of node k, Q k represents the reactive power of node k, P Gk represents the active power of the power generation node k, P Lk Represents the active power of load node k, V k represents the voltage amplitude of node k, V l represents the voltage amplitude of node l, G kl represents the conductance between node k and node l, B kl represents the susceptance between point k and node l, θ kl represents the voltage phase angle between point k and node l, Q Gk represents the reactive power of the generating node k, Q Lk represents the reactive power of load node k;

[0024] The node voltage and phase angle values ​​are derived separately to obtain the Jacobian matrix, and then the correction equation is solved according to the Jacobian matrix to obtain the voltage amplitude change ΔV k and the phase angle change Δδ k , and then obtain the new iterative node voltage amplitude and phase angle value.

[0025] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, the correction calculation includes:

[0026] The reactive power correction equation for the PV node is calculated as:

[0027] Q t+1 =Q t+f(ΔV t )

[0028] Among them, Q t+1 Indicates the reactive power value obtained after correction, Q t represents the reactive power value of the tth iteration, ΔV t represents the difference in voltage amplitude at the t-th iteration;

[0029] The calculation of the reactive power correction equation of the PQ(V) type node is:

[0030] Q t+1 =f(V t )

[0031] Among them, V t represents the voltage amplitude at the tth iteration;

[0032] The calculation of the reactive power correction equation of the PI type node is:

[0033]

[0034] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, wherein: the dynamic state estimation of the distribution network includes:

[0035] Based on various possible topologies resulting from different situations, the dynamic state estimation recurrent neural network is trained in advance using various data required for distribution network dynamic state estimation obtained through the digital twin model;

[0036] After obtaining the real-time status of the distribution network from the digital twin model, different prediction models are selected according to the topological status of the distribution network;

[0037] The vector Input the trained recurrent neural network to predict the node voltage amplitude and phase angle of the next sampling period, that is, in, Represents the active power vector of each power generation node at this moment, Represents the reactive power vector of each power generation node at this moment, Represents the active load vector of each load node at this moment, Represents the reactive load vector of each load node at this moment, Indicates the voltage amplitude of each node at this moment, Represents the phase angle vector of each node at this moment, Topo i Indicates the distribution network topology at this moment, Represents the prediction vector of the voltage amplitude of each node in the next sampling period, Represents the phase angle prediction vector of each node in the next sampling period.

[0038] As a preferred solution of the distribution network neural network dynamic state estimation method based on digital twins described in the present invention, it includes:

[0039] The calculation of the recurrent neural network is:

[0040] O t =g(WH t )

[0041] H t =f(UX t +VH t-1 )

[0042] Among them, O t represents the output value of the output layer at time t, g represents the activation function, W represents the weight matrix of the output layer, H t Represents the output value of the hidden layer at time t, H t-1 represents the output value of the hidden layer at time t-1, f represents the activation function, U represents the weight matrix of the input value, X t Represents the input value at time t, and V represents the value of the hidden layer at the previous moment as the weight matrix of the partial input at this moment;

[0043] By continuously bringing the calculation of the hidden layer into the calculation of the output layer, the output value of the recurrent neural network can be known as t Affected by the previous input values, a state C needs to be added to save the long-term dynamic state;

[0044] Using input data X t and h t-1 After the linear transformation, the result of the linear transformation is passed to the sigmoid function, and the sigmoid function maps the result of the linear transformation to the interval (0, 1) to control the flow of data and further realize the training of the recurrent neural network.

[0045] A second aspect of an embodiment of the present invention provides a distribution network neural network dynamic state estimation system based on digital twins, comprising:

[0046] The data acquisition unit is used to use each acquisition node in the actual distribution network to collect data at a pre-set acquisition cycle and transmit the data collected by each node to the control server via power carrier or wireless network communication;

[0047] A data processing unit, configured to filter and process the real-time sampled data using the digital twin model run by the control server, and to obtain various types of data required for dynamic state estimation of the distribution network through the digital twin model;

[0048] The state estimation unit is used to perform dynamic state estimation on the distribution network by using a recurrent neural network trained in advance according to the various types of data.

[0049] According to a third aspect of an embodiment of the present invention, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to call instructions stored in the memory to execute the steps of the method described in any embodiment of the present invention.

[0050] According to a fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, including:

[0051] When the computer program instructions are executed by a processor, the steps of the method according to any embodiment of the present invention are implemented.

[0052] Compared with the existing technology, the present invention has the following beneficial effects: the distribution network neural network dynamic state estimation method and system based on digital twins can filter and process raw distribution network data to obtain the accurate real-time state of the distribution network, and based on this, perform dynamic state estimation with good forecasting effect on high-proportion new energy distribution networks; it has a significant effect on the dispatch center obtaining the real-time state of the distribution network, pre-regulating distribution network risks, and effectively reducing the impact of new energy fluctuations. Compared with traditional Kalman filtering, it has faster computing speed and robustness, can adaptively switch to different prediction models according to different topological structures, and does not require a precise distribution network parameter model, which determines that it can have a good predictive effect on dynamic state estimation in new energy scenarios such as photovoltaics. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 The overall flow chart of the distribution network neural network dynamic state estimation method based on digital twin provided by the present invention;

[0055] Figure 2 A digital twin model correction data flow chart for the digital twin-based distribution network neural network dynamic state estimation method provided by the present invention;

[0056] Figure 3 The voltage support vector diagram of the distribution network neural network dynamic state estimation method based on digital twin provided by the present invention;

[0057] Figure 4A voltage drop scenario classification diagram for the distribution network neural network dynamic state estimation method based on digital twins provided by the present invention;

[0058] Figure 5 A flow chart of current reference value calculation for the distribution network neural network dynamic state estimation method based on digital twins provided by the present invention;

[0059] Figure 6 Schematic diagram of the IEEE 33-node distribution network of the distribution network neural network dynamic state estimation method based on digital twin provided by the present invention;

[0060] Figure 7 A comparison chart of the predicted and actual voltage amplitudes at node 19 in scenario 1 of the distribution network neural network dynamic state estimation method and system based on digital twins provided by the present invention;

[0061] Figure 8 A comparison chart of the predicted and actual voltage amplitudes at node 24 in scenario 1 of the distribution network neural network dynamic state estimation method and system based on digital twins provided by the present invention;

[0062] Figure 9 A comparison chart of the predicted and actual voltage amplitudes under impedance changes at the 19th node in scenario 1 of the distribution network neural network dynamic state estimation method and system based on digital twins provided by the present invention;

[0063] Figure 10 A comparison chart of the predicted and actual voltage amplitudes at node 19 with and without historical data in scenario 3 of the distribution network neural network dynamic state estimation method and system based on digital twins provided by the present invention;

[0064] Figure 11 A comparison chart of the predicted and actual voltage amplitudes of the 19th node distribution network photovoltaic access in scenario 4 of the distribution network neural network dynamic state estimation method and system based on digital twins provided by the present invention. DETAILED DESCRIPTION

[0065] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0066] Example 1

[0067] Reference Figures 1 to 5 In one embodiment of the present invention, a method for estimating the dynamic state of a distribution network neural network based on digital twins is provided, which specifically includes the following steps:

[0068] S1: Obtain real-time sampled data from the actual distribution network and send it to the control server.

[0069] It should be noted that:

[0070] Utilize each collection node in the actual distribution network to collect data in a pre-set collection cycle and transmit the data collected by each node to the control server using power carrier and wireless network communication methods.

[0071] S2: The digital twin model running on the control server filters and processes the real-time sampled data, and obtains various data required for dynamic state estimation of the distribution network through the digital twin model.

[0072] Given the power system network topology, component parameters, and generation and load parameters, the distribution of active power, reactive power, and voltage within the distribution network can be determined using limited distribution network measurement data to determine the steady-state parameters of the power system. Within the distribution network model, the distribution network state estimation algorithm can obtain various parameters of unmeasured points, including the voltage amplitude and phase angle at each busbar node, the power distribution of each branch, and the power loss of the network. While extracting the measurement data for calculation, the weighted least squares method can be used to filter and correct the raw data.

[0073] Furthermore, the DG model in the distribution network parameter calculation can basically be equivalent to four types: PV, PQ, PI and PQ(V) type nodes, as shown in Table 1.

[0074] Table 1: Distributed new energy model table.

[0075] technology Output Grid interface type type Node processing type wind power generation AC Direct grid connection Small, medium PQ node or PQ(V) node Photovoltaic power generation DC Inverter Small and medium PQ, PI or PV node micro gas turbine AC Direct grid connection Large PQ, PI or PV node Fuel power generation DC Inverter Medium and large PQ or PV node

[0076] Furthermore, the forward and backward parameter estimation algorithm is used to filter and process the real-time sampled data. The steps are as follows: Figure 2 As shown;

[0077] ① Based on the line impedance or admittance in the real-time sampling data obtained from the actual distribution network, the original node admittance matrix Y of the distribution network is formed, and the maximum number of iterations k is set;

[0078] It should be noted that the original node admittance matrix Y of the distribution network is expressed as:

[0079]

[0080] ② Select the balance node voltage and assign the voltage initial value V0=1 to other PI nodes in the whole network, and set the phase angle to 0. For PV type nodes, set the reactive initial value to 0; for PQ(V) type nodes, set the reactive initial value to Q0=f(V0); for PI type nodes, set the reactive initial value to Where I represents the current injected into the PI node, and P represents the active power injected into the PI node;

[0081] ③ Use the Newton-Raphson method to solve the voltage of each node and obtain the new iterative node voltage amplitude and phase angle value;

[0082] It should be noted that the Newton-Raphson method is used to solve the voltage of each node and calculate the active power and reactive power, which can be expressed as:

[0083]

[0084]

[0085] Among them, P k represents the active power of node k, Q k represents the reactive power of node k, P Gk represents the active power of the power generation node k, P Lk Represents the active power of load node k, V k represents the voltage amplitude of node k, V l represents the voltage amplitude of node l, G kl represents the conductance between node k and node l, B kl represents the susceptance between point k and node l, θ kl represents the voltage phase angle between point k and node l, Q Gk represents the reactive power of the generating node k, Q Lk represents the reactive power of load node k;

[0086] The node voltage and phase angle values ​​are derived to obtain the Jacobian matrix, and then the correction equation is solved according to the Jacobian matrix to obtain the voltage amplitude change ΔV k and the phase angle change Δδ k , and then obtain the new iterative node voltage amplitude and phase angle value.

[0087] ④ For PV type nodes, PQ(V) type nodes and PI type nodes, correction is performed according to their respective reactive power correction equations;

[0088] It should be noted that the calculation of the reactive power correction equation for PV type nodes is:

[0089] Q t+1 =Q t +f(ΔV t)

[0090] Among them, Q t+1 Indicates the reactive power value obtained after correction, Q t represents the reactive power value of the tth iteration, ΔV t represents the difference in voltage amplitude at the t-th iteration;

[0091] It should be noted that the calculation of the reactive power correction equation for the PQ(V) type node is:

[0092] Q t+1 =f(V t )

[0093] Among them, V t represents the voltage amplitude at the tth iteration;

[0094] It should be noted that the calculation of the reactive power correction equation of the PI type node is:

[0095]

[0096] ⑤Verify whether the convergence conditions are met based on the correction results;

[0097] It should be noted that the convergence conditions for PQ nodes, PQ(V) nodes, and PI nodes are:

[0098]

[0099] It should be noted that the convergence conditions for PV type nodes are:

[0100]

[0101] in, represents the voltage amplitude of node i at the k+1th iteration, represents the voltage amplitude of node i at the kth iteration, ε represents the preset threshold, V s Indicates the node voltage amplitude of the PV node.

[0102] ⑥ If the convergence conditions are met, the calculation ends and the number of iterations, convergence accuracy, voltage amplitude and phase are output. Otherwise, the calculation continues by jumping to the step of solving the voltage of each node.

[0103] It should be noted that through the above steps, the disturbed values ​​and missing values ​​in the original distribution network data can be corrected and supplemented in the process of multiple iterative calculations, thereby obtaining an accurate real-time distribution network state, which provides a data foundation for dynamic state estimation.

[0104] S3: The dynamic state of the distribution network is estimated through a recurrent neural network trained in advance based on various data. It should be noted that:

[0105] Dynamic state estimation of distribution network includes:

[0106] Based on various possible topologies resulting from different situations, the dynamic state estimation recurrent neural network is trained in advance using various data required for distribution network dynamic state estimation obtained through the digital twin model;

[0107] After obtaining the real-time status of the distribution network from the digital twin model, different prediction models are selected based on the topological status of the distribution network;

[0108] The vector Input the trained recurrent neural network to predict the node voltage amplitude and phase angle of the next sampling period, that is, in, Represents the active power vector of each power generation node at this moment, Represents the reactive power vector of each power generation node at this moment, Represents the active load vector of each load node at this moment, Represents the reactive load vector of each load node at this moment, Indicates the voltage amplitude of each node at this moment, Represents the phase angle vector of each node at this moment, Topo i Indicates the distribution network topology at this moment, Represents the prediction vector of the voltage amplitude of each node in the next sampling period, Represents the phase angle prediction vector of each node in the next sampling period. The steps are as follows Figure 5 shown.

[0109] Further, such as Figure 3 The left side of the figure shows a typical recurrent neural network. X is a vector representing the value of the input layer; H represents the vector of the hidden layer, U represents the weight matrix from the input layer to the hidden layer, O represents the vector of the output layer, W represents the weight matrix from the hidden layer to the output layer, and V represents the weight matrix where the value of the hidden layer at the previous moment is used as part of the input at this moment. If the recurrent neural network is expanded, a more intuitive framework can be obtained, namely Figure 3 On the right, the network receives input x at time t t , at this time the value of the hidden layer is H t , the output value is O t The difference from ordinary neural networks is that H t The value of X does not only depend on t , also depends on H t-1 ;

[0110] Furthermore, the calculation of the recurrent neural network is:

[0111] O t =g(WH t)

[0112] H t =f(UX t +VH t-1 )

[0113] Among them, O t represents the output value of the output layer at time t, g represents the activation function, W represents the weight matrix of the output layer, H t Represents the output value of the hidden layer at time t, H t-1 represents the output value of the hidden layer at time t-1, f represents the activation function, U represents the weight matrix of the input value, X t Represents the input value at time t, and V represents the value of the hidden layer at the previous moment as the weight matrix of the partial input at this moment;

[0114] By continuously bringing the calculation of the hidden layer into the calculation of the output layer, we can know the output value O of the recurrent neural network t Affected by previous input values, namely:

[0115] O t =g(WH t )

[0116] =Vf(UX t +VH t-1 )

[0117] =Vf(UX t +Vf(UX t-1 +VH t-2 ))

[0118] …

[0119] =Vf(UX t +Vf(UX t-1 +Vf(UX t-2 +Vf(UX t-3 +…))))

[0120] It should be noted that the recurrent neural network has the problem of gradient disappearance when facing long time series, that is, starting from time t-3, in most cases X t-3 The coefficients of the previous input layer are close to 0. That is, the network state H before time t-3 will not affect the output of the output layer during training, nor will it affect the update of the weight matrix W. The neural network has actually ignored the state before time t-3.

[0121] Therefore, the original RNN cannot handle long-distance data dependencies. Therefore, LSTM is used to estimate the dynamic state of the distribution network. The hidden layer of the original RNN has only one state, H, which is extremely sensitive to short-term effects. LSTM, on the other hand, adds a state C to store long-term state.

[0122] It should be noted that the idea of ​​LSTM is to use three gate switches. The first gate is responsible for controlling the continuation of the long-term state C; the second gate is responsible for controlling the input of the immediate state into the long-term state C; the third gate is responsible for controlling whether the long-term state C is used as the output of the current LSTM, such as Figure 4 The calculation of these three gates is based on the input data X t and h t-1 After the linear transformation, the result of the linear transformation is passed to the sigmoid function. The sigmoid function maps the result of the linear transformation to the interval (0,1) to control the data flow and thus realize the training of the recurrent neural network.

[0123] From the above, the beneficial effects of the present invention are:

[0124] The digital twin-based distribution network neural network dynamic state estimation method and system provided by this invention can filter and process raw distribution network data to obtain the accurate real-time state of the distribution network. Based on this, it can perform dynamic state estimation with good predictive effect for distribution networks with a high proportion of new energy. This method is of great significance for the dispatch center to obtain the real-time state of the distribution network, pre-regulate distribution network risks, and effectively reduce the impact of new energy fluctuations. Compared with traditional Kalman filtering, it has faster computing speed and robustness, can adaptively switch to different prediction models according to different topologies, and does not require a precise distribution network parameter model, which determines its good predictive effect for dynamic state estimation in new energy scenarios such as photovoltaics.

[0125] The second aspect of the present invention provides a distribution network neural network dynamic state estimation system based on digital twins, comprising:

[0126] The data acquisition unit is used to use each acquisition node in the actual distribution network to collect data at a pre-set acquisition cycle and transmit the data collected by each node to the control server via power carrier or wireless network communication;

[0127] The data processing unit is used to control the digital twin model running on the server to filter and process the real-time sampled data, and obtain various data required for dynamic state estimation of the distribution network through the digital twin model;

[0128] The state estimation unit is used to estimate the dynamic state of the distribution network through a recurrent neural network trained in advance based on various types of data.

[0129] According to a third aspect of the present invention, a computer device is provided, comprising a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication), or other technologies. When the computer program is executed by the processor, a method for synchronizing files between a terminal device and a carrier module is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a key, trackball, or touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse.

[0130] A fourth aspect of the present disclosure provides a computer-readable storage medium having computer program instructions stored thereon, including:

[0131] When the computer program instructions are executed by a processor, any of the above methods is implemented.

[0132] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0133] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0134] Example 2

[0135] Reference Figures 6 to 11 This is the second embodiment of the present invention. Different from the first embodiment, this embodiment provides a verification test of the distribution network neural network dynamic state estimation method and system based on digital twins to verify and illustrate the technical effects adopted in this method.

[0136] Simulation background: This embodiment verifies the dynamic state estimation of the distribution network neural network under the digital twin. The IEEE33-node distribution network model is used as the basis for simulation on RTDS (Real Time Digital Simulator). The schematic diagram of the IEEE33-node distribution network is shown in Figure 6 In different simulation scenarios, 3000 data sampling cycles are collected for each simulation, of which the first 80% are used as training sets and the last 20% are used as test sets.

[0137] Scenario 1: Dynamic state estimation of traditional distribution network. In this simulation scenario, no new energy power generation equipment is integrated into the distribution network. The simulation comparison diagram is as follows: Figure 7 and Figure 8 As shown in the figure, this simulation adjusts the simulation parameters based on the load conditions of the actual distribution network on the IEEE33-node distribution network load data.

[0138] Scenario 2: This simulation scenario simulates the increase in power line impedance in summer to test the accuracy of dynamic state estimation of distribution network under different meteorological conditions. The simulation results are as follows: Figure 9 As shown in the figure, this simulation adjusts the simulation parameters based on the load conditions of the actual distribution network on the IEEE33-node distribution network load data.

[0139] Scenario 3: Due to the fact that neural networks can accurately reflect historical data similar to the training set, this simulation scenario simulates a situation where the distribution network load has not been "remembered" by the neural network. This is used to test the accuracy of the distribution network dynamic state estimation in a situation where similar data has not appeared in the training set. The simulation comparison diagram is shown below. Figure 10 As shown in the figure, this simulation adjusts the simulation parameters based on the load conditions of the actual distribution network on the IEEE33-node distribution network load data.

[0140] Scenario 4: This simulation scenario simulates the situation after photovoltaic power generation equipment is connected to the distribution network at nodes 18 and 22 respectively. It is used to test the accuracy of the dynamic state estimation of the distribution network after the new energy power generation equipment is connected to the distribution network. The simulation comparison diagram is as follows: Figure 11This simulation uses the IEEE 33-node distribution network load data to adjust the simulation parameters based on the actual distribution network load conditions, and the photovoltaic parameters are adjusted using the regional total photovoltaic power generation power from the Belgian power grid company Elia.

[0141] It can be seen from the above simulation results that the method provided by the present invention solves the problems of interference and missing data transmission and collection, and is capable of screening and completing the required accurate data as the basic data for dynamic state estimation; compared with traditional Kalman filtering, it has faster calculation speed and robustness, can adaptively change to different prediction models according to different topological structures, and can have a better prediction effect on dynamic state estimation in new energy scenarios such as photovoltaics.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A distribution network neural network dynamic state estimation method based on digital twins, characterized by: include: Obtain real-time sampled data from the actual distribution network and send it to the control server; The digital twin model run by the control server filters and processes the real-time sampled data, and obtains various types of data required for dynamic state estimation of the distribution network through the digital twin model; The dynamic state of the distribution network is estimated by using a recurrent neural network trained in advance according to the various types of data.

2. The method for dynamic state estimation of a distribution network neural network based on digital twins according to claim 1, characterized in that: Utilize each collection node in the actual distribution network to collect data in a pre-set collection cycle and transmit the data collected by each node to the control server using power carrier and wireless network communication methods.

3. The method for dynamic state estimation of a distribution network neural network based on digital twins according to claim 2, characterized in that: Screening and processing the real-time sampled data includes: According to the line impedance or admittance in the real-time sampling data obtained from the actual distribution network, the original node admittance matrix Y of the distribution network is formed, and the maximum number of iterations k is set; Select the balance node voltage and assign the voltage initial value V0=1 to other PI nodes in the whole network, and set the phase angle to 0. For PV type nodes, set the reactive initial value to 0; for PQ(V) type nodes, set the reactive initial value to Q0=f(V0); for PI type nodes, set the reactive initial value to Where I represents the current injected into the PI node, and P represents the active power injected into the PI node; The Newton-Raphson method is used to solve the node voltage and obtain the new iterative node voltage amplitude and phase angle value. For PV type nodes, PQ (V) type nodes and PI type nodes, corrections are made according to their respective reactive power correction equations. Verify whether the convergence condition is met according to the correction result; The convergence conditions for PQ nodes, PQ(V) nodes, and PI nodes are: The convergence conditions for PV type nodes are: in, represents the voltage amplitude of node i at the k+1th iteration, represents the voltage amplitude of node i at the kth iteration, ε represents the preset threshold, V s Indicates the node voltage amplitude of the PV node; If the convergence condition is met, the calculation ends and the number of iterations, convergence accuracy, voltage amplitude and phase are output; otherwise, the calculation jumps to the step of solving the voltage of each node and continues.

4. The method for estimating the dynamic state of a distribution network neural network based on digital twins according to claim 3, characterized in that: The acquisition of the node voltage amplitude and phase angle value of the new iteration includes: The Newton-Raphson method is used to solve the voltage of each node and calculate the active power and reactive power, which can be expressed as: Among them, P k represents the active power of node k, Q k represents the reactive power of node k, P Gk represents the active power of the power generation node k, P Lk Represents the active power of load node k, V k represents the voltage amplitude of node k, V l represents the voltage amplitude of node l, G kl represents the conductance between node k and node l, B kl represents the susceptance between point k and node l, θ kl represents the voltage phase angle between point k and node l, Q Gk represents the reactive power of the generating node k, Q Lk represents the reactive power of load node k; The node voltage and phase angle values ​​are derived separately to obtain the Jacobian matrix, and then the correction equation is solved according to the Jacobian matrix to obtain the voltage amplitude change ΔV k and the phase angle change Δδ k , and then obtain the new iterative node voltage amplitude and phase angle value.

5. The method for dynamic state estimation of a distribution network neural network based on digital twins according to claim 3, characterized in that: The calculation of the correction includes, The reactive power correction equation for the PV node is calculated as: Q t+1 =Q t +f(ΔV t ) Among them, Q t+1 Indicates the reactive power value obtained after correction, Q t represents the reactive power value of the tth iteration, ΔV t represents the difference in voltage amplitude at the t-th iteration; The calculation of the reactive power correction equation of the PQ(V) type node is: Q t+1 =f(V t ) Among them, V t represents the voltage amplitude at the tth iteration; The calculation of the reactive power correction equation of the PI type node is:

6. The method for dynamic state estimation of a distribution network neural network based on digital twins according to claim 4 or 5, characterized in that: The dynamic state estimation of the distribution network includes: Based on various possible topologies resulting from different situations, the dynamic state estimation recurrent neural network is trained in advance using various data required for distribution network dynamic state estimation obtained through the digital twin model; After obtaining the real-time status of the distribution network from the digital twin model, different prediction models are selected according to the topological status of the distribution network; The vector Input the trained recurrent neural network to predict the node voltage amplitude and phase angle of the next sampling period, that is, in, Represents the active power vector of each power generation node at this moment, represents the reactive power vector of each power generation node at this moment, Represents the active load vector of each load node at this moment, Represents the reactive load vector of each load node at this moment, Indicates the voltage amplitude of each node at this moment, Represents the phase angle vector of each node at this moment, Topp i Indicates the distribution network topology at this moment, Represents the prediction vector of the voltage amplitude of each node in the next sampling period, Represents the phase angle prediction vector of each node in the next sampling period.

7. The method for estimating the dynamic state of a distribution network neural network based on digital twins according to claim 6, characterized in that: include, The calculation of the recurrent neural network is: O t =g(WH t ) H t =f(UX t +VH t-1 ) Among them, O t represents the output value of the output layer at time t, g represents the activation function, W represents the weight matrix of the output layer, H t Represents the output value of the hidden layer at time t, H t-1 represents the output value of the hidden layer at time t-1, f represents the activation function, U represents the weight matrix of the input value, X t Represents the input value at time t, and V represents the value of the hidden layer at the previous moment as the weight matrix of the partial input at this moment; By continuously bringing the calculation of the hidden layer into the calculation of the output layer, the output value of the recurrent neural network can be known as t Affected by the previous input values, a state C needs to be added to save the long-term dynamic state; Using input data X t and h t-1 After the linear transformation, the result of the linear transformation is passed to the sigmoid function, and the sigmoid function maps the result of the linear transformation to the interval (0, 1) to control the flow of data and further realize the training of the recurrent neural network.

8. A system for implementing the distribution network neural network dynamic state estimation method based on digital twin according to any one of claims 1 to 7, characterized in that: include: The data acquisition unit is used to use each acquisition node in the actual distribution network to collect data at a pre-set acquisition cycle and transmit the data collected by each node to the control server via power carrier or wireless network communication; A data processing unit, configured to filter and process the real-time sampled data using the digital twin model run by the control server, and to obtain various types of data required for dynamic state estimation of the distribution network through the digital twin model; The state estimation unit is used to perform dynamic state estimation on the distribution network by using a recurrent neural network trained in advance according to the various types of data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: The processor is configured to call the instructions stored in the memory to execute the steps of any one of the methods of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images