A digital twin model construction method for power grid dynamic security assessment and decision

Abstract

The application discloses a kind of digital twin model construction methods for power grid dynamic security assessment and decision-making, belong to electric power system control field, this method is based on including physical entity, virtual entity, twin data, connection, service five-dimensional framework constructs the digital twin model of power grid dynamic security assessment and decision-making, using service dimension to expand power grid digital twin application.When the current power system is running, the data of the next few rounds can be run in advance by the physical information neural network model, and the simulation results are output to the dynamic security assessment and decision-making service. Finally, the safety analysis results are output. According to the safety analysis results, power system operators can take measures in advance to manage risks, reduce potential safety hazards, improve the ability of power grid to handle faults and predict safety risks, and provide guidance for safe operation of power grid.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of power system control, and more particularly relates to a digital twin model construction method for power grid dynamic security assessment and decision-making. BACKGROUND

[0002] With the large-scale grid connection of renewable energy, the operation of the power grid is continuously approaching its safety limit, which may cause potential risks to the safe operation of the power grid. Dynamic security assessment (DSA) and decision-making are important means to ensure the safe and stable operation of the power grid. Under the current operating state, a set of possible contingencies are assessed for steady-state and dynamic security, and if the power grid is unstable under a certain contingency, appropriate preventive control measures are applied to correct the power grid operating state, thereby improving the security of the power grid. Existing dynamic security assessment and dynamic security decision-making are usually based on time-domain simulation methods, which require solving high-dimensional differential equations, resulting in long simulation time and difficulty in timely sensing potential security risks in the power grid. In order to improve the security perception and operation capability of the power grid, it is necessary to combine digital twin with power grid dynamic security assessment and decision-making, and simulate the future operation in the digital twin during the operation of the power grid, so as to perceive potential risks and problems in the operation of the power grid in advance and give corresponding optimization strategies and solutions.

[0003] Digital twin (DT) is a simulation technology that fully utilizes physical entities, sensor updates, operation history, etc. data to depict the actual behavior and state of physical entities in the real environment. Through virtual-real interaction, data fusion analysis, decision optimization, etc., the analysis and control optimization of physical entities, as well as the self-iterative optimization and updating of virtual entities, are realized. As a digital space model that can reflect the objective change law of physical objects, digital twin can predict the change trend of power grid operation and simulate and screen different security control strategies, thereby ensuring the security of the power grid.

[0004] However, the existing power grid dynamic security assessment and decision is usually based on a three-dimensional digital twin architecture (physical entity, virtual entity and connection), which is difficult to cope with the problems encountered in the process of power grid digital twin application, such as difficult effective data integration and processing, lack of scalability, difficult to introduce real-time decision, analysis, communication and other services. Secondly, the virtual entity in the existing power grid digital twin is generally based on pure model-driven or pure data-driven, and there are few virtual entities driven by mixed model and data, and there is also a lack of corresponding data-driven updating technology. Finally, the existing power grid dynamic security assessment process and decision process are usually considered separately: if the assessment is performed first and the decision is performed later, the adjusted operating state may have new effective contingency, leading to system instability; if iteration is performed between the assessment and decision processes, the respective software needs to exchange data through files, consuming program execution time. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a digital twin model construction method for power grid dynamic security assessment and decision, which is used for predicting the change trend of power grid operation, performing safety assessment on the operating state of the power grid, and correcting the operating state of the power grid through security constrained economic dispatch for unsafe conditions, so as to realize intelligent management and control of the power grid safety.

[0006] To achieve the above-mentioned purpose, according to the first aspect of the present application, a digital twin model construction method for power grid dynamic security assessment and decision is provided, comprising:

[0007] constructing a digital twin model for power grid dynamic security assessment and decision;

[0008] The digital twin model comprises five dimensions, namely physical entity, virtual entity, twin data, connection and service;

[0009] The physical entity is the power grid and its dynamic security assessment and decision process;

[0010] The virtual entity comprises:

[0011] a physical information neural network model for simulating and predicting the dynamic operation process of the power grid, taking the state variables, algebraic variables and control variables of the power grid in the steady state as inputs, and taking the state variables and algebraic variables of the power grid at the predicted time as outputs;

[0012] a security constrained economic dispatch model, taking the minimum generation cost of the power grid as the objective function, taking the generation capacity of the unit as the decision variable, and the constraint conditions including net load constraint, node power flow balance constraint and upper and lower limit constraint of unit generation capacity;

[0013] The service comprises:

[0014] S1, initializing the to-be-screened set S with the full set of contingenciesC , initializing the set of effective contingency S with an empty set A ;

[0015] S2, solving the security-constrained economic dispatch model to obtain the generation capacity that enables the power grid to operate normally when any of the contingencies in S A occurs;

[0016] S3, taking any contingency in S C as a target contingency, and determining whether the target contingency violates steady-state security, if yes, going to S4, otherwise going to S6;

[0017] S4, determining whether the target contingency violates dynamic security, if yes, using the physical information neural network model to predict the state variables of the power grid, substituting them into the dynamic process constraints and dynamic security constraints, and adding them into the constraint conditions of the security-constrained economic dispatch model; if no, adding the steady-state security constraints into the constraint conditions of the security-constrained economic dispatch model;

[0018] S5, deleting the target contingency from S C and adding it into S A , and going to S2;

[0019] S6, determining whether all contingencies in S C have been traversed, if yes, outputting the generation capacity that enables the power grid to operate normally when any of the contingencies in S A occurs, which is calculated by S2, if no, going to S3.

[0020] Preferably, the dynamic process constraints are:

[0021] The dynamic security constraints are: h(x(t), P g )≤0, t=0,...,t end ;

[0022] wherein x(t) is the state variable of the power grid at time t, P g is the generation capacity of generator g under normal conditions, g={1, 2,...G}, G is the total number of generators, M is a unit matrix, t end is the end time of the dynamic process of the power grid, and f(·) is a function of the generator dynamics model and h(·) is a function of the security metric;

[0023] The steady-state security constraints are transmission line capacity constraints.

[0024] Preferably, the determination of whether the target contingency violates dynamic security comprises:

[0025] The fault information of the target expected fault and steady-state flow information of the power grid are input into a pre-trained dynamic security assessment model based on a neural network to obtain a dynamic security state of the power grid, so as to determine whether the target expected fault violates dynamic security.

[0026] Preferably, the fault information includes: a fault element, a fault clearing element and a fault duration;

[0027] The steady-state flow information of the power grid includes: generator active power, generator reactive power, transmission line active power, transmission line reactive power, load active power, load reactive power and bus voltage amplitude and bus phase angle.

[0028] Preferably, the loss function of the physical information neural network model is:

[0029] wherein, D is the number of training data sets, and x j are the jth predicted value and the true value, respectively; and are the mean square error loss function, the differential regularization loss function and the physical loss function, respectively, and λ x , λ dt and λ f are hyperparameters in the corresponding loss function.

[0030] Preferably, the state variables include the angle and speed of the generator;

[0031] The algebraic variables include the bus voltage of the power grid;

[0032] The control variable is the power generation of the unit.

[0033] Preferably, the net load constraint is: d i =D i -r i ;

[0034] wherein, i={1, 2,..., I}, I is the total number of nodes of the power grid, D i is the load on node i, r i is the renewable energy generation on node i, and d i is the net load on node i;

[0035] The node flow balance constraint is:

[0036] wherein, for a transmission line l, the starting and ending nodes are α(l) and β(l) respectively, Φ(g) is the node where the unit g is located, is the total power generation of all generators at node i, is the power flow output from node i under the expected fault c, is the power flow to node i under the expected fault c,

[0037] The upper and lower limits of the unit power generation are:

[0038] Among them, p g is the power generation of generator set g, They are the lower and upper limits of power generation when the unit is turned on.

[0039] Preferably, the transmission line capacity constraint is:

[0040] Where l = {1, 2, ..., L}, c = {0, 1, 2, ..., L}, L is the total number of power transmission lines, c = 0 represents the normal situation, c = 1, 2, ..., L represents the expected circuit breaker fault of power transmission lines 1, 2, ..., L respectively, P l,c represents the power of transmission line l under the expected fault c, X l is the impedance of the transmission line l, represents the maximum transmission power of the transmission line l under the expected fault c, θ α(l),c ,θ β(l),c They represent the voltage phase angles at the nodes outputting and receiving power flows on the transmission line l under the expected fault c.

[0041] According to a second aspect of the present invention, a digital twin model construction device for dynamic power grid security assessment and decision-making is provided, comprising:

[0042] A processing module for building a digital twin model for dynamic grid security assessment and decision-making;

[0043] The digital twin model includes five dimensions: physical entity, virtual entity, twin data, connection and service;

[0044] The physical entity is the power grid and its dynamic security assessment and decision-making process;

[0045] The virtual entity includes:

[0046] Physical information neural network model, used to simulate and predict the dynamic operation process of the power grid, with the state variables, algebraic variables and control variables of the power grid in steady state as input, and the state variables and algebraic variables of the power grid at the predicted time as output;

[0047] The security constrained economic dispatch model takes the minimum generation cost of the units in the power grid as an objective function, takes the generation amount of the units as decision variables, and the constraint conditions include the net load constraint, the node power flow balance constraint, and the upper and lower limit constraints of the generation amount of the units;

[0048] The services include:

[0049] S1, initializing the set to be screened S from the full set of expected faults C , initializing the effective expected fault set S from an empty set A ;

[0050] S2, solving the security constrained economic dispatch model to obtain the generation amount that enables the power grid to operate normally when any effective expected fault in S A occurs;

[0051] S3, taking any expected fault in S C as a target expected fault, and judging whether the target expected fault violates the steady-state security, if yes, turning to S4, otherwise, turning to S6;

[0052] S4, if yes, using the physical information neural network model to predict the state variable of the power grid, substituting the state variable into the dynamic process constraint and the dynamic security constraint, and adding the state variable into the constraint conditions of the security constrained economic dispatch model; if no, adding the steady-state security constraint into the constraint conditions of the security constrained economic dispatch model;

[0053] S5, deleting the target expected fault from S C and adding the target expected fault into S A , and turning to S2;

[0054] S6, judging whether all expected faults in S C are traversed, if yes, outputting the generation amount that enables the power grid to operate normally when any effective expected fault in S A occurs, if no, turning to S3.

[0055] According to a third aspect of the present application, an electronic device is provided, comprising: a computer readable storage medium and a processor;

[0056] The computer readable storage medium is configured to store executable instructions;

[0057] The processor is configured to read the executable instructions stored in the computer readable storage medium, and execute the method according to the first aspect.

[0058] Overall, compared with the prior art, the above technical solutions conceived by the present application can achieve the following beneficial effects:

[0059] 1. The method provided by the present application is based on a five-dimensional framework including physical entities, virtual entities, twin data, connections and services to build a digital twin model of power grid dynamic security assessment and decision-making, solving the data and service related problems that the three-dimensional framework cannot handle. Through the twin data dimension of the five-dimensional framework, effective data integration and processing are carried out, and the service dimension is used to expand the power grid digital twin application, and services such as decision-making, analysis and communication are introduced. When the current power system is running, the data of the next few rounds can be run in the simulation model in advance, the simulation results are output to the dynamic security assessment and decision-making service, and finally the safety analysis results are output. According to the safety analysis results, the power system controller can take measures in advance to manage the risk and reduce potential safety hazards. Therefore, the digital twin model provided by the present application can simulate the power grid operation in advance, improve the ability of the power grid to handle faults and predict safety risks, and provide guidance for safe operation of the power grid.

[0060] 2. The method provided by the present application is based on a physical information neural network to model the dynamic operation process of the power grid, uses physical information to guide the data-driven model learning, and combines an incremental learning algorithm to update the model in real time, speed up the simulation time, and improve the prediction accuracy; compared with the pure model-driven method, the calculation speed is faster, and compared with the pure data-driven method, the calculation accuracy is higher.

[0061] 3. The method provided by the present application organically combines the dynamic security assessment process and the control decision process, proposes an integrated dynamic security assessment and decision-making algorithm, can iteratively solve the power grid dynamic security assessment and decision-making process in memory, and quickly screen the effective contingency set. In view of the unsafe situation, corresponding measures can be taken in time for prevention and control. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a five-dimensional digital twin architecture diagram of power grid dynamic security assessment and decision-making provided by an embodiment of the present application.

[0063] Figure 2 is a power grid dynamic process simulation model diagram based on a physical information neural network provided by an embodiment of the present application.

[0064] Figure 3 is an incremental learning diagram of a virtual entity provided by an embodiment of the present application.

[0065] Figure 4 is a dynamic security assessment service diagram based on a neural network provided by an embodiment of the present application.

[0066] Figure 5 is an integrated dynamic security assessment and decision-making service diagram based on a contingency screening algorithm provided by an embodiment of the present application.

[0067] Figure 6 is an application schematic diagram of power grid dynamic security assessment and decision based on digital twinning provided by an embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0069] An embodiment of the present application provides a digital twinning model construction method for power grid dynamic security assessment and decision, comprising:

[0070] constructing a digital twinning model for power grid dynamic security assessment and decision;

[0071] The digital twinning model comprises five dimensions, namely physical entity, virtual entity, twinning data, connection and service;

[0072] The physical entity is the power grid and its dynamic security assessment and decision process;

[0073] The virtual entity comprises:

[0074] a physical information neural network model for simulating and predicting the dynamic operation process of the power grid, taking the state variables, algebraic variables and control variables of the power grid under steady state as inputs, and taking the state variables and algebraic variables of the power grid at the predicted time as outputs;

[0075] a security constrained economic dispatch model, taking the minimum unit generation cost of the power grid as the objective function, taking the unit generation capacity as the decision variable, and the constraint conditions including net load constraint, node power flow balance constraint and upper and lower limit constraint of unit generation capacity;

[0076] The service comprises:

[0077] S1, initializing the to-be-screened set S C initializing the effective contingency set S A with an empty set;

[0078] S2, solving the security constrained economic dispatch model to obtain the generation capacity that enables the power grid to operate normally when any effective contingency in S A occurs;

[0079] S3, taking any contingency in S C as a target contingency, and determining whether the target contingency violates steady-state security, if yes, turning to S4, otherwise, turning to S6;

[0080] S4, if yes, the state variable of the power grid is predicted by using the physical information neural network model, which is substituted into the dynamic process constraint and the dynamic security constraint, and added to the constraint condition of the security constrained economic dispatch model; if no, the steady-state security constraint is added to the constraint condition of the security constrained economic dispatch model;

[0081] S5, the target contingency is deleted from S C and added to S A , and S2 is entered again;

[0082] S6, it is judged whether all contingencies in S C are traversed, if yes, the generation capacity calculated by S2, which enables the power grid to operate normally when any effective contingency in S A occurs, is output, if no, S3 is entered again.

[0083] Specifically, the method provided by the application comprises: obtaining power grid structure parameters and historical operation data; wherein the power grid structure parameters comprise generation cost, upper and lower limits of generation capacity of a generator unit, impedance and capacity of a power transmission line, and topological structure of the power grid.

[0084] According to the above data, a five-dimensional digital twin framework for power grid dynamic security assessment and decision-making is established; wherein the five-dimensional digital twin framework comprises five dimensions of physical entity, virtual entity, twin data, connection and service.

[0085] Then, actual power grid data is collected from the physical entity, and after being processed by the twin database, the data is input to the virtual entity for simulation, and the simulation results are processed by the digital twin service, and the power grid security analysis results are output.

[0086] The physical entity, the virtual entity, the twin data, the connection and the service are described below.

[0087] 1. Physical entity

[0088] The physical entity is the power grid and the dynamic security assessment and decision-making process thereof.

[0089] 2. Virtual entity

[0090] The virtual entity comprises two parts: a physical information neural network model and a security constrained economic dispatch model. The physical information neural network model is used to simulate and predict the dynamic operation process of the power grid. The security constrained economic dispatch model is used to depict the dispatching process of the power grid. The physical information neural network model and the security constrained economic dispatch model are described below.

[0091] The construction process of the physical information neural network model is as follows:

[0092] The dynamic process of the power grid can be described by the following equation set:

[0093]

[0094] g(x(t),y(t),u)=0, (2)

[0095] x(t0)=x0, (3)

[0096] y(t0)=y0, (4)

[0097] where t represents time. x(t) is a state variable of the power grid at time t, for example, the angle and speed of a generator. y(t) is an algebraic variable of the power system, for example, the total bus voltage of the power grid. u is a control variable of the power system, for example, the power generated. t0 is the initial time of the dynamic process, and x0 and y0 are initial values of the corresponding state variable and algebraic variable, respectively. f(·) and g(·) are functions describing the dynamic process of the power grid. Equation (1) and equation (2) are differential equations and algebraic equations describing the dynamic process, respectively, and equations (3) and (4) are initial value equations.

[0098] For the convenience of subsequent description, the system algebraic variable y(t) is incorporated into x(t) to represent, and a diagonal matrix M is used to distinguish the differential equation and the algebraic equation, so that the algebraic equation is incorporated into the differential equation to represent, at this time, the above equation (1) to equation (2) can be rewritten as:

[0099]

[0100] Figure 2 The power grid dynamic process simulation model based on the physical information neural network provided by the embodiment of the present application is a schematic diagram for data-driven modeling of the power grid dynamic process, and the basic network architecture of the physical information neural network model is a neural network model with K hidden layers (K is a positive integer greater than 1) as follows:

[0101]

[0102] where t is the prediction time, z0 is the input of the physical information neural network; W k , b k are the weight matrix and the bias vector of the kth layer of the physical information neural network, respectively, and σ is a nonlinear activation function, which is defined as is the output of the neural network at the prediction time t.

[0103] The loss function of the physical information neural network is as follows:

[0104]

[0105] where D is the number of historical training data sets, and x j represent the jth predicted value and true value, respectively. and are the mean square error loss function, differential regularization loss function and physical loss function, respectively, and λ x , λ dt and λ f are hyperparameters in the corresponding loss function. Training the above physical information neural network model can be performed by gradient descent method, using the above loss function to adjust the physical information neural network parameters.

[0106] The objective of the security constrained economic dispatch model is to minimize the unit generation cost, and its objective function is:

[0107]

[0108] where g = {1, 2,..., G} is the generator index, G is the total number of generators, C g (·) represents the generation cost function of unit g, p g represents the generation of generator g under normal conditions.

[0109] Further, the constraint conditions include: net load constraint, node power flow balance constraint, unit generation upper and lower limit constraint, steady-state security constraint (transmission line capacity constraint), dynamic process constraint, dynamic security constraint.

[0110] The net load constraint is:

[0111]

[0112] where i = {1, 2,..., I} is the node index, I is the total number of grid nodes, D i represents the load on node i, r i represents the renewable energy generation on node i, d i is the net load of node i.

[0113] The node power flow balance constraint is:

[0114]

[0115] where, represents the total generation of all generator units on node i, represents the power flow output from node i under contingency c, represents the power flow input to node i under contingency c.

[0116] The unit generation upper and lower limit constraint is:

[0117]

[0118] where p g represents the power generation of generator g, represents the lower and upper limits of power generation when the unit is on, respectively.

[0119] The transmission line capacity constraint is:

[0120]

[0121] where l={1,2,...,L} represents the transmission line index, L is the total number of transmission lines in the power grid, c={0,1,2,...,L} represents the index of the contingency, when c=0 represents the normal situation, and c=1,2,...,L represents the transmission line outage contingency. P l,c represents the power of transmission line l under contingency c, X l is the impedance of transmission line l. represents the maximum transmission power of transmission line l under contingency c. θ α(l),c , θ β(l),c represent the voltage phase angle of the node outputting and receiving power flow on transmission line l under contingency c, respectively.

[0122] The initial values of x0, y0 and decision variables p g are taken as the initial values of the steady-state process, and the dynamic process constraint under a certain contingency state can be obtained according to the above power grid dynamic process equation:

[0123]

[0124] where t end is the end time of the power grid dynamic process.

[0125] The dynamic stability constraint under a certain contingency state is:

[0126] h(x(t),P g )≤0 t=0,...,t end , (19)

[0127] where h(·) is the dynamic period related variable bound and inequality constraint.

[0128] 3. Twin data

[0129] The twin data module is used to collect and store data from physical entities and virtual entities, as well as manual operation data, and perform data processing, data fusion and other functions on the stored data.

[0130] Specifically, on the one hand, the twin data module collects real-time data from the power grid through measuring devices such as sensors, data acquisition and monitoring control systems, and wide-area measurement systems, and stores it in the operation database and dynamic safety database. Among them, the operation database includes historical data, planning data, and online data; the dynamic safety database includes anticipated faults, steady-state power flows, and corresponding safety tags. On the other hand, the twin data module can simulate and analyze the power grid through virtual entities to simulate the power grid state and collect data that is difficult to collect from physical entities. At the same time, the twin data module has data processing, analysis, and fusion functions. For example, it can fill in missing values ​​in the data, perform data enhancement on unbalanced data, and extract information from multi-dimensional data through fusion.

[0131] 4. Connect

[0132] Connections are the bridge and link for information exchange and sharing in digital twins, enabling interconnection and interoperability across all dimensions of digital twins. These connections include those between physical entities and twin data (CN_PD), physical entities and virtual entities (CN_PV), physical entities and services (CN_PS), virtual entities and twin data (CN_VD), virtual entities and services (CN_VS), and services and twin data (CN_SD).

[0133] Specifically, CN_PD can realize data collection and command issuance, and use sensors and other measuring devices in the power grid to collect real-time data of physical entities. The processed data and instructions in the corresponding twin data can be fed back to the physical entities to achieve power grid operation optimization. Figure 3For the incremental learning of the virtual entity of the embodiment of the present application, CN_PV, on the one hand, performs initialization training on various data-driven models in the virtual entity through historical data, so as to obtain an initial model with better performance. In actual operation, CN_PV can also realize incremental updating of the virtual entity through an incremental learning algorithm according to physical entity data collected by sensors, so as to correct output errors of the data-driven models in the virtual entity caused by problems such as changes in data distribution. On the other hand, data such as simulation analysis results of the virtual entity can be converted into control instructions and executed on the physical entity to realize real-time control of the power grid. The implementation of CN_PS is similar to that of CN_PD. Data collected by the power grid are transmitted to the service to realize updating and correction of dynamic security assessment and decision-making of the power grid, and results such as operation guidance, professional analysis and decision optimization generated by the service are provided to the power grid control center. CN_VD realizes interaction between the virtual entity and the twin data. On the one hand, simulation and related data generated by the virtual entity are stored in the twin data in real time, and on the other hand, high-quality analysis and processing data are extracted from the twin data to help modeling of the virtual entity. CN_VS realizes bidirectional communication between the virtual entity and the service, completes instruction delivery, data transmission and information synchronization of dynamic security assessment and decision-making. CN_SD is similar to CN_VD. On the one hand, data in the service are stored in the twin data in real time, and on the other hand, historical data, rule data, commonly used algorithms and models in the twin data are read in real time to support operation of dynamic security assessment and decision-making.

[0134] 5. Service

[0135] The service includes dynamic security assessment service and dynamic security decision service, and service encapsulation of various data, models and algorithms in the digital twin application.

[0136] The dynamic security assessment is used for identifying the dynamic security state of the power grid under a certain expected fault. The existing technology can be used for dynamic security assessment. In order to improve the assessment speed and accuracy, preferably, the dynamic security assessment service adopted by the present application is a dynamic security assessment service based on neural network, that is, a dynamic security assessment model based on neural network is used for dynamic security assessment service.

[0137] Figure 4 For the dynamic security assessment service based on neural network of the embodiment of the present application, the dynamic security assessment is used for identifying the dynamic security state of the power grid under a certain expected fault. The expected fault data, steady-state flow data and security labels required for constructing the dynamic security assessment model are extracted from the dynamic security database of the twin data, and data processing is performed thereon.

[0138] The expected fault and steady-state flow information are taken as features, and the corresponding security state is taken as a label to construct a training data set.

[0139] Feature = {ele f , ele c , t f , P G , Q G , P Line , Q Line , P L , Q L , V mB , V aB}, (20)

[0140] Label = {0, 1}, (21)

[0141] wherein, the contingency information includes: fault element ele f , fault clearing element ele c , fault duration t f . The steady state power flow information includes: engine active power P G , engine reactive power Q G , transmission line active power P Line , transmission line reactive power Q Line , load active power P L , load reactive power Q L , bus voltage amplitude V mB , bus phase angle V aB .

[0142] According to the training data set, a neural network model with N hidden layers is constructed to learn the mapping from features to labels. The dynamic security assessment model based on neural network is as follows:

[0143] Y0 = Feature, (22)

[0144]

[0145] wherein, Y0 is the input of the dynamic security assessment neural network; W n , b n are the weight matrix and bias vector of the nth layer of the neural network respectively, and σ(·) and sigmoid(·) are nonlinear activation functions, and their definitions are respectively sigmoid(v) = 1 / (1 + e -v ), is the predicted value of the dynamic security assessment neural network.

[0146] That is, the dynamic security assessment model of the neural network takes the contingency information of the contingency and the steady state power flow information of the power grid as samples, and trains to obtain the dynamic security state of the power grid as labels.

[0147] The loss function of the neural network-based dynamic security assessment model is:

[0148]

[0149] wherein Y is a real label, and the dynamic security sample of the power grid is defined as a positive sample, and the unsafe sample is defined as a negative sample. The neural network-based dynamic security assessment model can be trained by using the gradient descent method to adjust the parameters of the dynamic security assessment model by using the loss function.

[0150] The dynamic security decision service is a dynamic security decision service based on the expected fault screening algorithm.

[0151] Figure 5 The integrated dynamic security assessment and decision service based on the expected fault screening algorithm is an embodiment of the present application, which is used to solve the security constrained economic dispatch model, and the solving process includes:

[0152] Step 1: initializing the to-be-screened set S with the expected fault set C ={1,2,...,L}, and initializing the effective expected fault set with an empty set

[0153] Step 2: solving the security constrained economic dispatch problem in the virtual entity containing S A , and obtaining the power generation p A that enables the power grid to normally operate under normal conditions or under the condition that any effective expected fault in S g occurs.

[0154] That is, in the first cycle, since S A is an empty set, the power generation p g under the condition that the power grid normally operates without fault is obtained; when the expected fault meeting the condition is added to S A , the power generation p A that enables the power grid to normally operate under normal conditions or under the condition that any effective expected fault in S g occurs is calculated through the second cycle, and p C is updated.

[0155] Step 3: checking the next expected fault in S C , and using the steady-state security checking method to check whether the steady-state security is violated, and if the steady-state security is violated, turning to step 4, otherwise, turning to step 6.

[0156] wherein the steady-state security checking is performed by direct current flow calculation according to formula (17), and whether the transmission capacity of the line exceeds the threshold value is checked. If the threshold value is exceeded, it means that it is unsafe, otherwise, it is safe.

[0157] Step 4: Using the dynamic security assessment model described above, check whether the expected fault that has violated steady-state security also violates dynamic security. If dynamic security is violated, the expected fault is a valid expected fault, use the physical information neural network in the virtual entity to predict the state variables of the power grid under the current expected fault and substitute them into the dynamic process constraints and dynamic security constraints, and add the corresponding dynamic process constraints and dynamic security constraints to the security constrained economic dispatch problem; otherwise, add static security constraints to the security constrained economic dispatch problem.

[0158] Step 5: Add the current expected fault c to S A

[0159] S A ← S A ∪ c , (26)

[0160] Remove the current expected fault c from S C

[0161] S C ← S C \c, (27)

[0162] Go to Step 2.

[0163] It can be understood that the fault added to the valid expected fault set is the fault that will affect the security of the power grid, and the corresponding dispatching decision needs to be given, that is, the power generation capacity that can ensure the normal operation of the power grid when the corresponding fault occurs is given.

[0164] Step 6: Check whether all expected faults have been screened, if yes, output the optimal power grid operating state containing the expected faults. Otherwise, go to Step 3.

[0165] Figure 6 is the application schematic diagram of the power grid dynamic security assessment and decision based on digital twinning provided by the embodiment of the present application, the data collected from the power grid is transmitted to the twin data dimension, after analysis and processing, it is transmitted to the virtual entity for simulation and prediction, and the simulation results are output to the dynamic security assessment and decision service, and finally the security analysis results are output.

[0166] ​​The traditional power system cannot predict the security failure that may occur in the next round of evaluation when performing dynamic security evaluation, but when the method provided by the application is applied to the dynamic security evaluation and decision of the power grid, the real-time data of the power grid is collected by the measuring devices such as sensors, data acquisition and monitoring control systems, wide-area strategy systems, and transmitted to the operation database and dynamic security database of the twin data dimension, and the data processed by the twin data dimension can be transmitted to the virtual entity to realize the simulation and prediction of possible failures and potential risks. When the power system is running, the data of the next few rounds can be run in the simulation model in advance, the simulation results are output to the dynamic security evaluation and decision service, and finally the security analysis results are output. According to the security analysis results, the power system controller can take measures to manage the risk in advance and reduce potential security risks.

[0167] The embodiment of the application provides a digital twin model construction device for power grid dynamic security evaluation and decision, comprising:

[0168] A processing module is configured to construct a digital twin model for power grid dynamic security evaluation and decision.

[0169] The digital twin model comprises five dimensions, namely physical entity, virtual entity, twin data, connection and service.

[0170] The physical entity is the power grid and its dynamic security evaluation and decision process.

[0171] The virtual entity comprises:

[0172] A physical information neural network model is configured to simulate and predict the dynamic operation process of the power grid, wherein the state variables, algebraic variables and control variables of the power grid in the steady state are input, and the state variables and algebraic variables of the power grid at the prediction time are output.

[0173] A security constrained economic dispatch model is configured to minimize the unit generation cost of the power grid as an objective function, and the unit generation capacity as a decision variable, and the constraint conditions include net load constraint, node power flow balance constraint and upper and lower limit constraint of unit generation capacity.

[0174] The service comprises:

[0175] S1, initializing the to-be-screened set S with the set of all expected failures C Initializing the effective expected failure set S with an empty set A ;

[0176] S2, solving the security constrained economic dispatch model to obtain the generation capacity that enables the power grid to operate normally when any effective expected failure in S A occurs;

[0177] S3, adding the effective expected failure set SC Any expected fault in is taken as a target expected fault, and whether the target expected fault violates steady-state safety is determined. If so, the process proceeds to S4; otherwise, the process proceeds to S6;

[0178] S4, inputting the fault information of the target anticipated fault and the steady-state power flow information of the power grid into a pre-trained dynamic safety assessment network to obtain the dynamic safety state of the power grid; if the target anticipated fault violates dynamic safety, using the physical information neural network model to predict the state variables of the power grid to substitute them into the dynamic process constraints and dynamic safety constraints, and adding them to the constraints of the security-constrained economic dispatch model; otherwise, adding the steady-state safety constraints to the constraints of the security-constrained economic dispatch model;

[0179] S5, the target expected fault is changed from S C Delete and add S A , transfer to S2;

[0180] S6, determine whether to traverse S C If all the expected faults in S are present, then the output is the one calculated by S2 that makes the power grid A The power generation capacity that can operate normally in the event of any valid expected fault in the system, otherwise it goes to S3.

[0181] An embodiment of the present invention provides an electronic device, comprising: a computer-readable storage medium and a processor;

[0182] The computer-readable storage medium is used to store executable instructions;

[0183] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the method described in any one of the above embodiments.

[0184] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the method described in any one of the above embodiments.

[0185] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a digital twin model for dynamic security assessment and decision-making of power grids, characterized by: include: Build a digital twin model for dynamic grid security assessment and decision-making; The digital twin model includes five dimensions: physical entity, virtual entity, twin data, connection and service; The physical entity is the power grid and its dynamic security assessment and decision-making process; The virtual entity includes: Physical information neural network model, used to simulate and predict the dynamic operation process of the power grid, with the state variables, algebraic variables and control variables of the power grid in steady state as input, and the state variables and algebraic variables of the power grid at the predicted time as output; The safety-constrained economic dispatch model takes minimizing the power generation cost of the power grid unit as the objective function and the power generation of the unit as the decision variable. The constraints include net load constraints, node power flow balance constraints, and upper and lower limits of power generation of the unit. The services include: S1, initialize the set to be screened S with the full set of expected faults C , initialize the valid expected fault set S with an empty set A ; S2, solve the security constraint economic dispatch model to obtain the power grid in the event of S A The amount of power generated that can be normally operated in the event of any valid anticipated failure; S3, S C Any expected fault in is taken as a target expected fault, and whether the target expected fault violates steady-state safety is determined. If so, the process proceeds to S4; otherwise, the process proceeds to S6; S4, determining that the target anticipated fault violates dynamic safety; if so, using the physical information neural network model to predict the state variables of the power grid, substituting the state variables into the dynamic process constraints and dynamic safety constraints, and adding them to the constraints of the safety-constrained economic dispatch model; if not, adding steady-state safety constraints to the constraints of the safety-constrained economic dispatch model; S5, the target expected fault is changed from S C Delete and add S A , transfer to S2; S6, determine whether to traverse S C If all the expected faults in S are present, then the output is the one calculated by S2 that makes the power grid A The power generation capacity that can operate normally in the event of any valid expected fault in the system, otherwise it goes to S3.

2. The method according to claim 1, wherein The dynamic process constraints are: The dynamic safety constraint is: h(x(t),P g )≤0,t=0,...,t end ; Among them, x(t) is the state variable of the power grid at time t, p g is the power generation of generator g under normal conditions, g={1,2,...G}, G is the total number of generators, M is the unit matrix, t end is the end time of the power grid dynamic process, f(·) is the generator dynamic model function, and h(·) is the function of security measurement; The steady-state safety constraint is a transmission line capacity constraint.

3. The method according to claim 1 or 2, wherein: The determining that the target anticipated fault violates dynamic safety includes: The fault information of the target anticipated fault and the steady-state power flow information of the power grid are input into a pre-trained dynamic safety assessment model based on a neural network to obtain the dynamic safety status of the power grid, so as to determine whether the target anticipated fault violates dynamic safety.

4. The method according to claim 3, wherein The fault information includes: fault component, fault clearing component and fault duration; The steady-state power flow information of the power grid includes: engine active power, engine reactive power, transmission line active power, transmission line reactive power, load active power, load reactive power and bus voltage amplitude and bus phase angle.

5. The method according to claim 1, wherein The loss function of the physical information neural network model is: in, D is the number of training data sets, and x j are the jth predicted value and true value respectively; and They are the mean square error loss function, differential regularization loss function and physical loss function, λ x ,λ dt and λ f are the hyperparameters in the corresponding loss functions.

6. The method according to claim 1, wherein The state variables include the angle and speed of the generator; The algebraic variables include grid bus voltage; The controlled variable is the power generation of the unit.

7. The method according to claim 1, wherein The net load constraint is: d i =D i -r i ; Where i = {1, 2, ..., I}, I is the total number of grid nodes, D i is the load on node i, r i is the renewable energy generation at node i, d i is the net load on node i; The node power flow balance constraint is: Among them, for the transmission line l, its starting and ending nodes are α(l) and β(l) respectively, Φ(g) is the node where the unit g is located, is the total power generation of all generators at node i, is the power flow output from node i under the expected fault c, is the power flow input to node i under the expected fault c, The upper and lower limits of the unit power generation are: Among them, p g is the power generation of generator set g, They are the lower and upper limits of power generation when the unit is turned on.

8. The method according to claim 1, wherein The transmission line capacity constraint is: Where l = {1, 2, ..., L}, c = {0, 1, 2, ..., L}, L is the total number of power transmission lines, c = 0 represents the normal situation, c = 1, 2, ..., L represents the expected circuit breaker fault of power transmission lines 1, 2, ..., L respectively, P l,c represents the power of transmission line l under the expected fault c, X l is the impedance of the transmission line l, represents the maximum transmission power of the transmission line l under the expected fault c, θ α(l),c ,θ β(l),c They represent the voltage phase angles at the nodes outputting and receiving power flows on the transmission line l under the expected fault c.

9. A digital twin model construction device for dynamic security assessment and decision-making of power grids, characterized by: include: A processing module for building a digital twin model for dynamic grid security assessment and decision-making; The digital twin model includes five dimensions: physical entity, virtual entity, twin data, connection and service; The physical entity is the power grid and its dynamic security assessment and decision-making process; The virtual entity includes: Physical information neural network model, used to simulate and predict the dynamic operation process of the power grid, with the state variables, algebraic variables and control variables of the power grid in steady state as input, and the state variables and algebraic variables of the power grid at the predicted time as output; The safety-constrained economic dispatch model takes minimizing the power generation cost of the power grid unit as the objective function and the power generation of the unit as the decision variable. The constraints include net load constraints, node power flow balance constraints, and upper and lower limits of power generation of the unit. The services include: S1, initialize the set to be screened S with the full set of expected faults C , initialize the valid expected fault set S with an empty set A ; S2, solve the security constraint economic dispatch model to obtain the power grid in the event of S A The amount of power generated that can be normally operated in the event of any valid anticipated failure; S3, S C Any expected fault in is taken as a target expected fault, and whether the target expected fault violates steady-state safety is determined. If so, the process proceeds to S4; otherwise, the process proceeds to S6; S4, if yes, predict the state variables of the power grid using the physical information neural network model, substitute the state variables into the dynamic process constraints and dynamic safety constraints, and add them to the constraints of the safety-constrained economic dispatch model; if no, add the steady-state safety constraints to the constraints of the safety-constrained economic dispatch model; S5, the target expected fault is changed from S C Delete and add S A , transfer to S2; S6, determine whether to traverse S C If all the expected faults in S are present, then the output is the one calculated by S2 that makes the power grid A The power generation capacity that can operate normally in the event of any valid expected fault in the system, otherwise it goes to S3.

10. An electronic device, characterized in that: include: Computer-readable storage medium and processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the method according to any one of claims 1 to 8.

Images