A method for preventing transient stability of power system based on deep cross-network

Abstract

The present invention relates to the field of power system automation technology, and discloses a power system transient stability prevention and control method based on a deep cross network, including: S1, in the offline stage, setting power system related parameters to form a power grid operation mode database; S2, setting a set of expected faults to form a transient stability database of the power system; S3, building a transient stability assessment model of the power system based on a deep cross network; S4, in the online stage, determining the power grid operation mode of the next scheduling period, and using the deep cross network model to predict the transient stability of the operation mode; S5, based on the deep cross network model, deriving the numerical sensitivity of the active power of each generator set in the power system to the transient stability margin of the power system, and sorting and screening the key generator sets; S6, building and solving an optimization decision model for the power grid operation mode of the key generator set containing transient stability constraints. The present invention effectively improves the safety level of power system operation.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and in particular to a power system transient stability preventive control method based on a deep cross network. Background Art

[0002] In order to prevent large-scale power outages caused by transient instability, effective online assessment of transient stability of power systems and preventive control decision-making technology are urgently needed. The traditional transient stability assessment method is a model-driven time-domain simulation method, that is, a dynamic model of a differential-algebraic equation group is constructed for the target power system, and the transient stability of the system is analyzed by time-domain simulation calculation. The model-driven preventive control method requires the construction of an optimal power flow model containing a differential-algebraic equation group and transient stability constraints, and the solution methods include direct and indirect methods. The direct method performs difference processing on the differential-algebraic equation group and transforms the original dynamic optimization problem into a large-scale nonlinear optimization problem, but it is extremely difficult to solve the large-scale nonlinear optimization model; the indirect method requires repeated time-domain simulation calculations in each iteration, so the solution takes a long time. Both of the above model-driven methods are difficult to meet the needs of online transient stability analysis and preventive control. Summary of the invention

[0003] The present invention provides a method for preventing and controlling transient stability of an electric power system based on a deep cross network, which realizes rapid evaluation of transient stability of an electric power system based on a deep cross network under anticipated fault constraints, establishes an artificial intelligence-driven optimization decision model for an electric power grid operation mode containing transient stability constraints and solves the model to obtain a preventive control strategy. The method can quickly predict the safe state of the electric power grid operation mode, issue early warnings for operation modes with risks of instability, and generate adjustment strategies for power generation plans. The adjusted operation modes meet the requirements for transient stability and safe operation of the electric power system.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for preventing and controlling transient stability of a power system based on a deep cross network, comprising:

[0006] S1. In the offline stage, power system related parameters are set, and a number of random power grid operation modes are generated based on Monte Carlo sampling according to the power system related parameters to form a power grid operation mode database;

[0007] S2. Setting an expected fault set, and combining any expected fault scenario in the expected fault set with any power grid operation mode in the power grid operation mode database to perform time domain simulation calculation and transient stability analysis, so as to form a transient stability database of the power system;

[0008] S3, constructing a transient stability assessment model of the power system based on a deep cross network, and using the transient stability database to complete parameter training of the transient stability assessment model to obtain several perfect transient stability assessment models based on a deep cross network;

[0009] S4, online stage, determine the grid operation mode for the next dispatch period, and use the deep cross-network model to predict the transient stability of the operation mode;

[0010] S5. Based on the deep cross network model, the numerical sensitivity of the active power of each generator set in the power system to the transient stability margin of the power system is derived, and the key generator sets involved in preventive control are sorted and selected;

[0011] S6. Construct an optimization decision model for the grid operation mode of the key generating units with transient stability constraints, and use the successive linear approximation method to solve the optimization decision model to obtain a preventive control strategy based on the active power optimization adjustment of the key generating units, so as to meet the grid operation mode required for stable operation and realize transient stability preventive control of the power system.

[0012] As an optimization, in S1, the power system related parameters include the active power and reactive power of the load and the fluctuation range of the active power of the generator set. According to the power system related parameters, a number of random power grid operation modes are generated based on Monte Carlo sampling to form a power grid operation mode database. The specific steps are:

[0013] S1.1. For the target power system, based on the day-ahead load forecast and the combination results of the generator sets, the benchmark operating conditions for a total of 96 time points in the next 24 hours are formed;

[0014] S1.2, define the mean of the active power and reactive power of the load as the central value of the predicted value, and randomly sample the active power and reactive power of the load according to a normal distribution with a standard deviation of a% to obtain a random value of the load demand, where a is a positive number;

[0015] S1.3, given an adjustable generator set for active power balance in real-time operation of the power system, the adjustable range of active power of the adjustable generator set is set to [P uc -20%P N ,P uc +20%P N ], where P uc is the active power output plan of the generator set generated by the generator set combination on the previous day, P N is the rated power of the generator set, and random sampling is performed according to uniform distribution within the active power range of each generator set to obtain a random value of the generator set output;

[0016] S1.4. Generate a number of random power grid operation modes through the Monte Carlo sampling method and perform power flow calculation, and save the power grid operation modes with converged power flow and satisfying the static safety operation rules into the power grid operation mode database.

[0017] As an optimization, in S2, the specific steps for forming a transient stability database are:

[0018] S2.1. Set an expected fault set for the target power system, and obtain the power grid operation mode database generated in step S1, select a power grid operation mode from the power grid operation mode database each time, and select a fault scenario from the expected fault set, and use a binary search method to calculate the critical removal time of the expected fault scenario;

[0019] S2.2. Traverse the combination of all power grid operation modes in the power grid operation mode database and all anticipated fault scenarios in the anticipated fault set, calculate the critical removal time of all anticipated fault scenarios, use the active power of the first generator set as the input feature, and the critical removal time of the anticipated fault scenario as the target attribute, and summarize all corresponding input features and target attributes to form a transient stability data knowledge base.

[0020] As an optimization, the specific steps of using the binary search method to calculate the critical removal time of the expected fault scenario are:

[0021] S2.1.1. Input the grid operation mode and expected fault scenario before the grid fault, and initialize the binary interval to [t lower ,t upper ]=[0,10],t lower ,t upper are the lower and upper bounds of the two partitions, respectively, and t lower ,t upper The unit of is cycle;

[0022] S2.1.2. Set the fault clearing time t c =t lower , perform time domain simulation on the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, judge that the power system is transiently unstable and enter S2.1.4. Otherwise, when δmax<360°, judge that the power system remains transiently stable and enter S2.1.3;

[0023] S2.1.3, according to the given step value t step Extend the fault removal time, and then refresh the binary interval, that is, assign the upper limit value of the original binary interval to the lower limit value of the new binary interval, and the upper limit value of the new binary interval is the upper limit value of the original binary interval plus the given step value, that is, t lower New = tupper Old, t upper New = t upper Old+t step , then return to S2.1.2;

[0024] S2.1.4. Reset the fault clearing time to t c =(t lower +t upper ) / 2, perform time domain simulation on the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, it is judged that the power system is transiently unstable. Otherwise, when δmax<360°, it is judged that the power system maintains transient stability.

[0025] S2.1.5. If the power system maintains transient stability, the fault clearing time t re-determined by S2.1.4 is used. c Replace the lower bound t between the two partitions lower A new binary interval is formed; if the power system is in a transient instability, the fault clearing time t re-determined by S2.1.4 is used c Replace the upper bound t between the two partitions upper A new binary interval is formed to refresh the binary interval;

[0026] S2.1.6. Check the span between the two partitions. If the span gap between the two partitions is t upper -t lower <1, then end the binary search and go to S2.1.7; otherwise, return to S2.1.4 to continue the binary search;

[0027] S2.1.7. Output critical clearing time of expected fault scenario CCT = t lower .

[0028] As an optimization, in S3, the specific steps of constructing a transient stability assessment model based on a deep cross network and completing parameter training of the transient stability assessment model using the transient stability database are:

[0029] Based on the transient stability database generated in step S2, the active power and excitation reference voltage of the generator set are used as input features, and the critical removal time of the expected fault scenario is used as the target attribute. A transient stability assessment model based on a deep cross network is constructed, and the transient stability database generated in step S2 is used to complete the parameter training of the transient stability assessment model. After the training is completed, the transient stability assessment model of the deep cross network corresponding to each expected fault scenario is stored in the rule base.

[0030] As an optimization, several transient stability assessment models based on deep cross networks generated in step S3 are expressed as:

[0031]

[0032] Among them, Φ C CCT is the transient stability assessment model of deep cross network, Φ C CCT The subscript C in the above indicates that the expected fault set contains C expected fault scenarios, and also indicates that S3 generates a total of C deep cross-network evaluation models to predict the critical removal time of the corresponding fault scenarios. CCT represents the critical removal time of the expected fault scenario, and P G is the active power of the generator, V G is the generator voltage, P L and Q L They represent the load active power and reactive power respectively.

[0033] As an optimization, in S4, the operation mode of the power grid in the next dispatch period is determined, and the specific method of using the deep cross network model to predict the transient stability of the operation mode is as follows:

[0034] S4.1. During the online operation phase of the power grid, the power grid operation mode for the next dispatching period is determined based on the short-term load forecast results and the day-ahead generator unit combination results;

[0035] S4.2. Use the deep cross network model to quickly estimate the critical removal time of each expected fault scenario of the power system under this operation mode. If the critical removal time CCT of any of the expected fault scenarios in the power system is less than the action time t set by its relay protection device, r , it is determined that the power system has a risk of transient instability, and the estimated result is returned to the grid operation and dispatching personnel, and an alarm signal is returned.

[0036] As an optimization, in S5, the numerical sensitivity of each generator's active power to the system transient stability margin is derived based on the deep cross network model, and the specific steps of sorting and selecting the key generators participating in preventive control are as follows:

[0037] S5.1, step S3 generates several perfect transient stability assessment models based on deep cross network, and calculates the total differential of the transient stability assessment model to the active power of each generator set as:

[0038]

[0039] In the formula, is the mathematical expansion of the deep cross network model for the critical removal time of the cth expected fault, N G is the total number of generating units, ΔP G，i Represents the change in active power of the i-th generator set;

[0040] S5.2, using the partial differential corresponding to the total differential obtained in S5.1 to measure the sensitivity of each generator set's active power to the critical removal time;

[0041] S5.3. For each expected fault scenario, calculate the sensitivity of all generator sets in the power system. If the sensitivity is positive, the generator set has a positive effect on the stability of the power system, and the active output of the generator set should be increased. On the contrary, if the sensitivity is negative, the generator set has a negative effect on the stability of the power system, and the active output of the generator set should be reduced.

[0042] S5.3. Sort the generator sets from large to small according to their sensitivity, select the generator sets with the top b% in the positive sensitivity ranking as the generator set that can increase the output, and select the generator sets with the bottom b% in the negative sensitivity ranking as the generator set that can reduce the output, where b is a positive number;

[0043] S5.4. Summarize the corresponding generator sets participating in output adjustment and the active output data of the generator sets under each expected fault scenario to form the generator sets for grid transient stability preventive control and the corresponding decision variables.

[0044] As an optimization, in S6, the optimization target of the optimization decision model is to minimize the economic cost of the active power adjustment of the key generator set, and the constraints include the grid DC power flow equation constraint, the transient stability inequality constraint, the line active power flow upper and lower limit safety constraints, and the feasibility inequality constraint of the generator active output adjustment. The expression of the optimization decision model is as follows:

[0045]

[0046]

[0047] in, and are respectively the positive adjustment amount and negative adjustment amount of the active power of the key generating set, is the active power of the key generator set in the initial operation mode, and are the unit cost coefficients for positive and negative adjustment of the active power of the generator set, G + and G - They represent the generator sets participating in the positive and negative regulation of active power, B is the grid node admittance matrix, and θ is the voltage phase vector of each node. represents a deep cross network model, and Both represent the sensitivity of the active power of the generator set to the critical removal time, t rIndicates the time when the relay protection device operates and cuts off the fault line. P ij and They represent the upper and lower limits of the branch active power flow, L represents the set of power grid branches,<i,j> ∈L means there is a branch between node i and node j, θ i and θ j They represent the voltage phases of node i and node j respectively, Indicates the maximum limit of the positive adjustment of the active power of the generator set. Indicates the maximum limit of negative adjustment of the active power of the generator set.

[0048] As an optimization, in S6, the specific steps of using the successive linear approximation method to solve the optimization decision model to obtain a power grid operation mode that meets the stable operation requirements are:

[0049] S6.1. Initialize the parameters of the optimization decision model to determine the active power of the key generator set under the initial operation mode. And set the initial operation mode as the current operation mode;

[0050] S6.2. using a deep cross-network to evaluate the critical removal time of each anticipated fault scenario under the current operation mode of the power grid, and estimating the sensitivity of the critical removal time to the active power of the key generator set;

[0051] S6.3, Substitute the sensitivity of the active power of the key generator set in the current operation mode into the optimization decision model, solve the optimization decision model to obtain the positive adjustment amount of the active power of the key generator set and the negative adjustment of the active power of the key generator set Correcting the active power of the key generator set and setting the new operation mode as the current operation mode;

[0052] S6.4. Use the deep cross network to evaluate the transient stability of the current operation mode of the power grid. If the power system can maintain transient stability for any expected fault scenario, proceed to step S6.5; otherwise, proceed to step S6.2;

[0053] S6.5. When the active power adjustment amount of the generator set is less than the preset threshold, the active power adjustment strategy of the key generator set is terminated and output, and a power grid operation mode that meets the stable operation requirements is obtained according to the adjustment strategy; otherwise, return to step S6.2.

[0054] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0055] Online transient stability assessment and preventive control are important means to prevent major power outages in power systems. The present invention proposes a method for preventing transient stability of power systems based on a deep cross network, realizes rapid transient stability assessment of power grids based on a deep cross network, establishes an operating mode optimization model with transient stability constraints driven by neural network rules and solves the model, and obtains a preventive control strategy based on the optimal adjustment of the active power of the generator set. This method can give early warnings for operating modes with risks of instability and generate preventive control strategies, effectively improving the safety level of power system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0057] Figure 1 A method flow chart of a method for preventing and controlling transient stability of a power system based on a deep cross network according to the present invention;

[0058] Figure 2 It is a schematic diagram of the structure of the deep cross network model;

[0059] Figure 3 It is a structural diagram of the IEEE 39-node test system of an embodiment of the present invention;

[0060] Figure 4 The active output diagram of each unit before and after preventive control using the method of the present invention;

[0061] Figure 5 The method of the present invention is used to predict the generator power angle trajectory under fault conditions before and after preventive control. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0063] A method for preventing and controlling transient stability of a power system based on a deep cross network, comprising:

[0064] S1. In the offline stage, power system related parameters are set, and a number of random power grid operation modes are generated based on Monte Carlo sampling according to the power system related parameters to form a power grid operation mode database.

[0065] In this embodiment, in S1, the power system related parameters include the active power and reactive power of the load and the fluctuation range of the active power of the generator set. According to the power system related parameters, a number of random power grid operation modes are generated based on Monte Carlo sampling to form a power grid operation mode database. The specific steps are:

[0066] S1.1. For the target power system, based on the day-ahead load forecast and the combination results of the generator sets, the benchmark operating conditions for a total of 96 time points in the next 24 hours are formed;

[0067] Day-ahead load forecasting is an important part of power grid operation. The present invention can use existing methods to complete day-ahead load forecasting and provide input data for the method of the present invention.

[0068] S1.2. Define the mean of the active power and reactive power of the load as the central value of the predicted value, and randomly sample the active power and reactive power of the load according to a normal distribution with a standard deviation of a% to obtain a random value of the load demand, where a is a positive number; considering that the typical error level of power load prediction is 5%, the standard deviation is defined as 5%.

[0069] S1.3, given an adjustable generator set for active power balance in real-time operation of the power system, the adjustable range of active power of the adjustable generator set is set to [P uc -20%P N ,P uc +20%P N ], where P uc is the active power output plan of the generator set generated by the generator set combination on the previous day, P N is the rated power of the generator set, and random sampling is performed according to uniform distribution within the active power range of each generator set to obtain a random value of the generator set output;

[0070] Similarly, the combination of generator sets on the day before is an important part of the operation of the power grid. The present invention does not involve proposing a completely new method for combining generator sets on the day before. The existing method can be used to complete the combination of generator sets on the day before and provide input data for the method of the present invention.

[0071] S1.4. Generate a number of random power grid operation modes through the Monte Carlo sampling method and perform power flow calculation, and save the power grid operation modes with converged power flow and satisfying the static safety operation rules into the power grid operation mode database.

[0072] The “random values ​​of load demand” and “random values ​​of generator output” are generated through the Monte Carlo sampling method.

[0073] The power flow converges and meets the static safety operation rules as follows:

[0074] First, regarding power flow convergence, power flow calculation relies on the comprehensive power system stability program PSASP (or other power system analysis software), and the calculation results of the program contain information on whether the power flow converges.

[0075] Secondly, regarding the "static security operation rules", it mainly analyzes whether the node voltage amplitude exceeds the limit. The node voltage amplitude should satisfy the safety range of 0.95 p.u. <= V <= 1.05 p.u. (p.u. means per unit value), and whether the current of the line and transformer (collectively referred to as branches) exceeds the limit. The branch current should satisfy I < Imax (Imax represents the rated capacity of the line and transformer itself). When both of the above two conditions are met, it is considered that the power grid operation mode satisfies the static security operation rules.

[0076] S2. Set a set of contingency faults, and perform time-domain simulation calculation and transient stability analysis by combining any one contingency fault scenario in the set of contingency faults with any one power grid operation mode in the power grid operation mode database, so as to form a transient stability database of the power system. That is, perform time-domain simulation calculation and transient stability analysis for any combination of "power grid operation mode - contingency fault scenario" to form a transient stability database.

[0077] In this embodiment, in S2, the specific steps for forming the transient stability database are as follows:

[0078] S2.1. Set a set of contingency faults for the target power system, and obtain the power grid operation mode database generated in step S1. Each time, select one power grid operation mode from the power grid operation mode database, and select one fault scenario from the set of contingency faults, and calculate the critical clearing time (CCT) of this contingency fault scenario using the binary search method.

[0079] In this embodiment, the specific steps for calculating the critical clearing time of the contingency fault scenario using the binary search method are as follows:

[0080] S2.1.1. Input the power grid operation mode before the power grid fault and the contingency fault scenario, and initialize the binary search interval as [t lower , t upper = [0, 10], where t lower , t upper are the lower and upper bounds of the binary search interval respectively, and the unit of t lower , t upper is cycle;

[0081] S2.1.2. Set the fault clearing time t c = t lower, perform time domain simulation on the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, judge that the power system is transiently unstable and enter S2.1.4. Otherwise, when δmax<360°, judge that the power system remains transiently stable and enter S2.1.3;

[0082] The power angle difference is the basis for judging whether the system is transiently stable or not. For the transient stability problem of the power system, the longer the fault removal time is, the more serious the disturbance of the power system is, and the worse the transient stability is. The binary interval is to search for the fault removal time of the critical state of the power system (i.e. the boundary between stability and instability).

[0083] For the expected fault, the power system comprehensive stability program PSASP is used to perform time domain simulation calculation. The PSASP program outputs the generator power angle trajectory data after the expected fault occurs in the target power system under the representative power grid flow mode, and the transient stability index (TSI) is calculated as follows:

[0084]

[0085]

[0086] In the formula, Δδ max is the maximum unit power angle difference during the time domain simulation, t represents the time variable, assuming that the expected fault occurs at the initial time (0 seconds), and the final time of the time domain simulation calculation is T and T = 20 seconds, is the generator set of the target power system, δ represents the generator power angle, and δ i (t) and δ j (t) refers to the power angle of the i-th generator and the j-th generator at the t-th second respectively.

[0087] When TSI>0, the target power system can maintain transient stability under a given expected fault; conversely, when TSI<0, the target power system will experience transient instability under a given expected fault.

[0088] S2.1.3, according to the given step value t step Extend the fault removal time, and then refresh the binary interval, that is, assign the upper limit value of the original binary interval to the lower limit value of the new binary interval, and the upper limit value of the new binary interval is the upper limit value of the original binary interval plus the given step value, that is, t lower New = t upper Old, t upper New = t upper Old+t step , then return to S2.1.2;

[0089] S2.1.4. Reset the fault clearing time to t c =(t lower +t upper ) / 2, perform time domain simulation on the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, it is judged that the power system is transiently unstable. Otherwise, when δmax<360°, it is judged that the power system maintains transient stability.

[0090] S2.1.5. If the power system maintains transient stability, the fault clearing time t re-determined by S2.1.4 is used. c Replace the lower bound t between the two partitions lower A new binary interval is formed; if the power system is in a transient instability, the fault clearing time t re-determined by S2.1.4 is used c Replace the upper bound t between the two partitions upper A new binary interval is formed to refresh the binary interval;

[0091] S2.1.6. Check the span between the two partitions. If the span gap between the two partitions is t upper -t lower <1, then end the binary search and go to S2.1.7; otherwise, return to S2.1.4 to continue the binary search;

[0092] S2.1.7. Output critical clearing time of expected fault scenario CCT = t lower .

[0093] S2.2. Traverse the combination of all power grid operation modes in the power grid operation mode database and all anticipated fault scenarios in the anticipated fault set, calculate the critical removal time of all anticipated fault scenarios, use the active power of the first generator set as the input feature, and the critical removal time of the anticipated fault scenario as the target attribute, and summarize all corresponding input features and target attributes to form a transient stability data knowledge base.

[0094] S3. Construct a transient stability assessment model for the power system based on a deep cross network, and use the transient stability database to complete parameter training of the transient stability assessment model to obtain several perfect transient stability assessment models based on a deep cross network.

[0095] In this embodiment, the specific steps of constructing a transient stability assessment model based on a deep cross network and completing parameter training of the transient stability assessment model using the transient stability database are as follows:

[0096] Based on the transient stability database generated in step S2, the active power and excitation reference voltage of the generator set are used as input features, and the critical removal time of the expected fault scenario is used as the target attribute. A transient stability assessment model based on a deep cross network is constructed, and the transient stability database generated in step S2 is used to complete the parameter training of the transient stability assessment model. After the training is completed, the transient stability assessment model of the deep cross network corresponding to each expected fault scenario is stored in the rule base.

[0097] The deep cross network (DCN) used in the present invention includes a cross network (CrossNetwork) and an output layer (Output Layer). The overall deep cross network model structure is as follows: Figure 2 shown.

[0098] 1. Cross-network

[0099] The feedforward calculation formula of the cross network is as follows:

[0100]

[0101] In the formula, are the output features of the lth layer and the l+1th layer respectively; is the weight of the lth layer, T represents the vector transpose, that is, the vector is the feature vector x of the lth layer l The transpose of .

[0102] 2. Output Layer

[0103] The output layer obtains the final prediction result through linear mapping:

[0104]

[0105] Where W out and b out are the weight and bias of the output layer, respectively. For the prediction results.

[0106] The loss function of the feed-back training of the deep cross network model is:

[0107]

[0108] In the formula, y i is the true label of the training sample, N is the total number of training samples, λ is the coefficient of L2 regularization, and w l Generally refers to the model parameters of each layer of the deep cross network. After setting the loss function, the Adam algorithm can be used to learn the parameters.

[0109] In this embodiment, the several transient stability assessment models based on deep cross network generated in step S3 are expressed as:

[0110]

[0111] Among them, Φ C CCT is the transient stability assessment model of deep cross network, Φ C CCT The subscript C in the above indicates that the expected fault set contains C expected fault scenarios, and also indicates that S3 generates a total of C deep cross-network evaluation models to predict the critical removal time of the corresponding fault scenarios. CCT represents the critical removal time of the expected fault scenario, and P G is the active power of the generator, V G is the generator voltage, P L and Q L They represent the load active power and reactive power respectively.

[0112] S4: In the online stage, the grid operation mode of the next scheduling period is determined, and the transient stability of the operation mode is predicted using the deep cross-network model.

[0113] Here, the following steps can be used to determine the grid operation mode for the next dispatch period:

[0114] First, the load demand for the next dispatching period is determined according to the load forecast (load forecasting is an important work link in the operation of the power grid. The present invention does not involve the innovation of the load forecasting method, and the existing power load forecasting method can be used);

[0115] Secondly, the generator startup status and active output of the next dispatching period are determined according to the day-ahead power generation plan (the day-ahead power generation plan setting is also an important work link of the power grid operation. The present invention does not involve the innovation of the relevant method of the day-ahead power generation plan, and the existing day-ahead power generation plan can be used)

[0116] After determining the load demand and the active output of the generator, the grid operation mode can be obtained through grid flow calculation. (Grid flow calculation is an existing and mature grid analysis and calculation method)

[0117] Specifically, in this embodiment, in S4, the specific method of determining the power grid operation mode of the next scheduling period and predicting the transient stability of the operation mode using the deep cross network model is:

[0118] S4.1, during the online operation stage of the power grid, the power grid operation mode of the next dispatch period is determined based on the short-term load forecast (i.e., the day-ahead load forecast in S1.1) and the day-ahead generator combination results. After the load forecast results and the generator combination results are determined, the power flow calculation of the power system is calculated to obtain the power grid operation mode, which is the power flow calculation result;

[0119] S4.2. Use the deep cross network model to quickly estimate the critical removal time of each expected fault scenario of the power system under this operation mode. If the critical removal time CCT of any of the expected fault scenarios in the power system is less than the action time t set by its relay protection device, r , it is determined that the power system has a risk of transient instability, and the estimated result is returned to the grid operation and dispatching personnel, and an alarm signal is returned.

[0120] The deep cross network model is used to quickly estimate the critical removal time of each expected fault scenario of the power system under this operation mode. Specifically, the deep cross network model generated by S3 is used to transform P G , V G , P L , Q L The input is fed into the deep cross network model, and the neural network feedforward operation of the model is completed to output the critical resection time.

[0121] The estimation result is the conclusion on whether there is a risk of transient instability in the power system.

[0122] S5. Based on the deep cross network model, the numerical sensitivity of the active power of each generator set in the power system to the transient stability margin of the power system is derived, and the key generator sets involved in preventive control are sorted and selected;

[0123] In this embodiment, in S5, the numerical sensitivity of each generator active power to the system transient stability margin is derived based on the deep cross network model, and the specific steps of sorting and screening the key generator sets participating in the preventive control are:

[0124] S5.1, step S3 generates several perfect transient stability assessment models based on deep cross network, and calculates the total differential of the transient stability assessment model to the active power of each generator set as:

[0125]

[0126] In the formula, is the mathematical expansion of the deep cross network model for the critical removal time of the cth expected fault, N G is the total number of generating units, ΔP G，i Represents the change in active power of the i-th generator set;

[0127] S5.2, using the partial differential corresponding to the total differential obtained in S5.1 to measure the sensitivity of each generator set's active power to the critical removal time;

[0128] S5.3. For each expected fault scenario, calculate the sensitivity of all generator sets in the power system. If the sensitivity is positive, the generator set has a positive effect on the stability of the power system, and the active output of the generator set should be increased. On the contrary, if the sensitivity is negative, the generator set has a negative effect on the stability of the power system, and the active output of the generator set should be reduced.

[0129] S5.3. Sort the generator sets from large to small according to sensitivity, select the generator sets ranked in the top b% in terms of positive sensitivity as the set of generator sets that can increase output, and select the generator sets ranked in the bottom b% in terms of negative sensitivity as the set of generator sets that can reduce output, where b is a positive number; here, b is 25, that is, the generator sets ranked in the top 25% in terms of positive sensitivity are selected as the set of generator sets that can increase output, and the generator sets ranked in the bottom 25% in terms of negative sensitivity are selected as the set of generator sets that can reduce output.

[0130] S5.4. Summarize the corresponding generator sets participating in output adjustment and the active output data of the generator sets under each expected fault scenario to form the generator sets for grid transient stability preventive control and the corresponding decision variables.

[0131] S6. Construct an optimization decision model for the grid operation mode of the key generating units with transient stability constraints, and use the successive linear approximation method to solve the optimization decision model to obtain a preventive control strategy based on the active power optimization adjustment of the key generating units, so as to meet the grid operation mode required for stable operation and realize transient stability preventive control of the power system.

[0132] In this embodiment, in S6, the optimization target of the optimization decision model is to minimize the economic cost of active power adjustment of the generator sets participating in preventive control, and the constraints include the grid DC power flow equation constraint, the transient stability inequality constraint, the line active power flow upper and lower limit safety constraints, and the feasibility inequality constraint of the generator active output adjustment. The expression of the optimization decision model is as follows:

[0133]

[0134]

[0135] in, and are respectively the positive adjustment amount and negative adjustment amount of the active power of the generator set participating in preventive control, is the active power of the generator set participating in preventive control in the initial operation mode, and are the unit cost coefficients for positive and negative adjustment of the active power of the generator set, G + and G - They represent the generator sets participating in the positive and negative regulation of active power, B is the grid node admittance matrix, and θ is the voltage phase vector of each node. represents a deep cross network model, and Both represent the sensitivity of the active power of the generator set to the critical removal time, t r Indicates the time when the relay protection device operates and cuts off the fault line. P ij and They represent the upper and lower limits of the branch active power flow, L represents the set of power grid branches,<i,j> ∈L means there is a branch between node i and node j, θ i and θ j They represent the voltage phases of node i and node j respectively, Indicates the maximum limit of the positive adjustment of the active power of the generator set. Indicates the maximum limit of negative adjustment of the active power of the generator set.

[0136] In this embodiment, in S6, the specific steps of using the successive linear approximation method to solve the optimization decision model to obtain a power grid operation mode that meets the stable operation requirements are:

[0137] S6.1. Initialize the parameters of the optimization decision model to determine the active power of the generator sets participating in the preventive control under the initial operation mode. And set the initial operation mode as the current operation mode;

[0138] S6.2. Use a deep cross-network to evaluate the critical clearing time of each anticipated fault scenario under the current operation mode of the power grid, and estimate the sensitivity of the critical clearing time to the active power of the generator sets involved in the regulation; the estimation can refer to the methods of steps S5.1 and S5.2.

[0139] S6.3, Substitute the sensitivity of the active power of the generator set participating in the regulation in the current operation mode into the optimization decision model, solve the optimization decision model to obtain the positive adjustment amount of the active power of the generator set participating in the regulation And the negative adjustment of the active power of the generator sets involved in the regulation Correct the active power of the generator sets involved in the regulation and set the new operation mode as the current operation mode;

[0140] S6.4. Use the deep cross network to evaluate the transient stability of the current operation mode of the power grid. If the power system can maintain transient stability for any expected fault scenario, proceed to step S6.5; otherwise, proceed to step S6.2. The specific evaluation process can refer to step S4.2: Use the deep cross network model to quickly estimate the critical removal time of each expected fault scenario of the power system under this operation mode. If the critical removal time CCT of any of the expected fault scenarios in the power system is less than the action time t set by its relay protection device, r , it is determined that the power system has a risk of transient instability.

[0141] S6.5. When the active power adjustment amount of the generator set is less than the preset threshold (such as 10MW), the active power adjustment strategy of the generator set involved in the regulation is terminated and output, and the power grid operation mode that meets the stable operation requirements is obtained according to the adjustment strategy, otherwise return to step S6.2.

[0142] The adjustment strategy refers to optimizing the values ​​of the parameters of the decision model, that is, and The "grid operation mode" here includes the "active power of the generator set" and the active power and reactive power of the load, etc. If the active power of the generator set is adjusted while the load remains unchanged, the result of the grid flow calculation will also change, that is, the "grid operation mode" will change.

[0143] Next, we take the IEEE 39-node test system as an example. Figure 3 shown.

[0144] The expected fault set is set as shown in Table 1. "*" indicates that a permanent three-phase short circuit fault occurs in the faulty line on the bus side. After 0.05 seconds, the protection device starts to cut off the faulty line and isolate the fault.

[0145] Table 1 Anticipated fault set

[0146]

[0147] The active output of the generators and the active / reactive demand of the loads of the whole network are adjusted proportionally to 9 load levels of 80%, 85%, ..., 120% times the reference value. The corresponding operation modes under different load levels are regarded as the predicted operation mode set generated based on the day-ahead load forecast and unit combination. The load active / reactive demand is set to fluctuate randomly with a normal distribution with a mean of the reference value and a variance of 5% times the reference value. The active output of the generator fluctuates randomly with a uniform distribution within the range of [Puc-20%PN, Puc+20%PN]. 5500 random operating conditions are generated based on Monte Carlo random sampling to form a set of power grid operating conditions.

[0148] Considering the expected fault set shown in Table 1, transient stability analysis is performed on all combinations of "grid operation mode-expected fault", the PSASP power system analysis comprehensive program is used for time domain simulation analysis, and the binary search method is used to solve the critical removal time of the expected fault under this operation mode. Traverse all grid operation conditions-expected fault combinations, record the active power of the generator set and the critical removal time of the expected fault under all operation modes, and generate a transient stability database.

[0149] The power grid operating condition set is randomly divided into a training set, a verification set, and a test set in a ratio of 60:20:20. A deep cross network is built based on the PyTorch library, and the training set data is used to train the model. The Adam algorithm is used for model training, and the learning rate is 0.001. The verification set is used to determine whether the deep cross network is overfitting. The generalization ability of the neural network model is verified and evaluated at each training generation. When the prediction accuracy of the verification set data gradually decreases and the prediction accuracy of the training set data gradually increases, the training is terminated in advance to avoid further model overfitting. The test set data is assumed to be the actual operating conditions of the real-time operation stage of the power grid, and is used in the present invention to demonstrate the prediction accuracy of the deep cross network for the transient stability assessment of the power grid. Table 2 gives the root mean square error RMSE of the deep cross network for the critical removal time of the expected fault. Taking fault scenario #1 as an example, the RMSE is 0.42 cycles. Since the prediction error obeys the normal distribution, according to the statistical characteristics of the normal distribution, 95% of the test samples have prediction errors falling within the interval of [-2×0.42,+2×0.42]. That is, the absolute value of the prediction error is not greater than 2×0.42=0.84 cycles, and the credibility is 95%.

[0150] Table 2 Root mean square error of deep cross network for critical removal time of expected fault

[0151] Failure scenario #1 #2 #3 RMSE / cycle 0.42 0.39 0.37

[0152] When the real-time operation of the power grid is judged to be unstable, preventive control should be carried out in time to avoid transient instability. Based on the deep cross network, the optimization model of power grid transient stability preventive control is constructed, and the preventive control strategy is solved by the successive linear approximation method. The active output of each unit before and after preventive control is shown in Figure 2. Figure 4 shown.

[0153] Finally, the effectiveness of the preventive control strategy is verified. The PSASP program is used to perform time domain simulation calculations and transient stability analysis on the expected fault set. The generator power angle trajectory of the power grid before and after the preventive control is as follows Figure 5 shown.

[0154] from Figure 5Comparison of the power angle trajectories before and after preventive control shows that under the original grid operating conditions, anticipated faults 1-3 can cause transient instability in the grid; after power generation redispatching preventive control, the grid can tolerate anticipated faults 1-3 under the new grid operating conditions. Therefore, after preventive control, the grid can restore transient stability and effectively mitigate the risk of transient instability.

[0155] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. 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 preventing and controlling transient stability of a power system based on a deep cross network, characterized in that: include: S1. In the offline stage, power system related parameters are set, and a number of random power grid operation modes are generated based on Monte Carlo sampling according to the power system related parameters to form a power grid operation mode database; S2. Setting an expected fault set, and combining any expected fault scenario in the expected fault set with any power grid operation mode in the power grid operation mode database to perform time domain simulation calculation and transient stability analysis, so as to form a transient stability database of the power system; S3, constructing a transient stability assessment model of the power system based on a deep cross network, and using the transient stability database to complete parameter training of the transient stability assessment model to obtain several perfect transient stability assessment models based on a deep cross network; The specific steps are: Based on the transient stability database generated in step S2, the active power and excitation reference voltage of the generator set are used as input features, and the critical removal time of the expected fault scenario is used as the target attribute. A transient stability assessment model based on a deep cross network is constructed, and the transient stability assessment model is trained using the transient stability database generated in step S2. After the training is completed, the transient stability assessment model of the deep cross network corresponding to each expected fault scenario is stored in the rule base; The generated transient stability assessment models based on deep cross networks are expressed as follows: ； in, is a transient stability assessment model for deep cross-networks. The subscript C in the above indicates that the expected fault set contains C expected fault scenarios, which also means that S3 generates a total of C A deep cross-network evaluation model is used to predict the critical removal time of the corresponding fault scenario. CCT represents the critical removal time of the expected fault scenario. is the generator active power, is the generator voltage, and Represent the load active power and reactive power respectively; S4, online stage, determine the grid operation mode for the next dispatch period, and use the deep cross-network model to predict the transient stability of the operation mode; The specific method is: S4.

1. During the online operation phase of the power grid, the power grid operation mode for the next dispatching period is determined based on the short-term load forecast results and the day-ahead generator combination results; S4.

2. Use the deep cross network model to quickly estimate the critical removal time of each expected fault scenario of the power system under this operation mode. If the critical removal time CCT of any of the expected fault scenarios in the power system is less than the set action time of its relay protection device, t r , it is determined that the power system has a risk of transient instability, and the estimated result is returned to the grid operation and dispatching personnel, and an alarm signal is returned; S5. Based on the deep cross network model, the numerical sensitivity of the active power of each generator set in the power system to the transient stability margin of the power system is derived, and the key generator sets involved in preventive control are sorted and selected; The specific steps are: S5.1, step S3 generates several perfect transient stability assessment models based on deep cross network, and calculates the total differential of the transient stability assessment model to the active power of each generator set as: ； In the formula, is the mathematical expansion of the deep cross network model for the critical removal time of the cth expected fault, is the total number of generating units, Indicates i The change in active power of each generator set; Indicates i Active power of generator sets; S5.2, using the partial differential corresponding to the total differential obtained in S5.1 to measure the sensitivity of each generator set's active power to the critical removal time; S5.

3. For each expected fault scenario, calculate the sensitivity of all generator sets in the power system. If the sensitivity is positive, the generator set has a positive effect on the stability of the power system, and the active output of the generator set should be increased. On the contrary, if the sensitivity is negative, the generator set has a negative effect on the stability of the power system, and the active output of the generator set should be reduced. S5.

3. Sort the generator sets from large to small according to their sensitivity, select the generator sets with the top b% in positive sensitivity as the generator set that can increase output, and select the generator sets with the bottom b% in negative sensitivity as the generator set that can reduce output, where b is a positive number; S5.

4. Summarize the corresponding generator sets participating in output adjustment and the active output data of the generator sets under each expected fault scenario to form the generator sets and corresponding decision variables for grid transient stability prevention control S6. Construct an optimization decision model for the grid operation mode of the key generating units with transient stability constraints, and use the successive linear approximation method to solve the optimization decision model to obtain a preventive control strategy based on the active power optimization adjustment of the key generating units, so as to meet the grid operation mode required for stable operation and realize transient stability preventive control of the power system.

2. The method for preventing and controlling transient stability of a power system based on a deep cross network according to claim 1 is characterized in that: In S1, the power system related parameters include the active power and reactive power of the load and the fluctuation range of the active power of the generator set. According to the power system related parameters, a number of random power grid operation modes are generated based on Monte Carlo sampling to form a power grid operation mode database. The specific steps are: S1.

1. For the target power system, based on the day-ahead load forecast and the combination results of the generator sets, a benchmark operating condition for a total of 4*h time points in the future h hours is formed, where h is a positive integer; S1.

2. Define the mean of the active power and reactive power of the load as the central value of the predicted value, and randomly sample the active power and reactive power of the load according to a normal distribution with a standard deviation of a% to obtain a random value of the load demand, where a is a positive number; S1.

3. Given an adjustable generator set for active power balance in real-time operation of the power system, set the adjustable active power range of the adjustable generator set to ,in, The active power output plan of the generator set generated by the generator set combination on the previous day, is the rated power of the generator set, and random sampling is performed according to uniform distribution within the active power range of each generator set to obtain a random value of the generator set output; S1.

4. Generate a number of random power grid operation modes through the Monte Carlo sampling method and perform power flow calculation, and save the power grid operation modes with converged power flow and satisfying the static safety operation rules into the power grid operation mode database.

3. The method for preventing and controlling transient stability of a power system based on a deep cross network according to claim 1 is characterized in that: In S2, the specific steps of forming the transient stability database are: S2.

1. Set an expected fault set for the target power system, and obtain the power grid operation mode database generated in step S1, select a power grid operation mode from the power grid operation mode database each time, and select a fault scenario from the expected fault set, and use a binary search method to calculate the critical removal time of the expected fault scenario; S2.

2. Traverse the combination of all power grid operation modes in the power grid operation mode database and all anticipated fault scenarios in the anticipated fault set, calculate the critical removal time of all anticipated fault scenarios, use the active power of the first generator set as the input feature, and the critical removal time of the anticipated fault scenario as the target attribute, and summarize all corresponding input features and target attributes to form a transient stability data knowledge base.

4. The method for preventing and controlling transient stability of a power system based on a deep cross network according to claim 3 is characterized in that: The specific steps of using the binary search method to calculate the critical removal time of the expected fault scenario are: S2.1.

1. Input the grid operation mode and expected fault scenario before the grid fault, and initialize the binary interval to , , are the lower and upper bounds of the two partitions, respectively. , The unit of is cycle; S2.1.

2. Setting the fault clearing time , perform time domain simulation on the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, judge that the power system is transiently unstable and enter S2.1.

4. Otherwise, when δmax<360°, judge that the power system remains transiently stable and enter S2.1.3; S2.1.3, according to the given step value t step Extend the fault removal time, and then refresh the binary interval, that is, assign the upper limit value of the original binary interval to the lower limit value of the new binary interval, and the upper limit value of the new binary interval is the upper limit value of the original binary interval plus the given step value, that is , , then return to S2.1.2; S2.1.

4. Reset the fault clearing time to , conduct time domain simulation of the expected fault scenario, and make stability judgment based on whether the power angle difference δmax of the largest generator set is greater than 360°. When δmax>360°, it is judged that the power system is transiently unstable. Otherwise, when δmax<360°, it is judged that the power system maintains transient stability. S2.1.5: If the power system maintains transient stability, the fault clearing time re-determined in S2.1.4 shall be used. t c Replace the lower bound between the two partitions A new binary interval is formed; if the power system is in a transient instability, the fault clearing time re-determined by S2.1.4 is used Replace the upper bound between the two partitions A new binary interval is formed to refresh the binary interval; S2.1.

6. Check the span between the two partitions. If the span gap between the two partitions is , then the binary search ends and the process goes to S2.1.7; otherwise, the process goes back to S2.1.4 and continues the binary search; S2.1.

7. Output critical clearing time of expected fault scenario .

5. The method for preventing and controlling transient stability of a power system based on a deep cross network according to claim 1, characterized in that: In S6, the optimization target of the optimization decision model is to minimize the economic cost of the active power adjustment of the key generator set, and the constraints include the grid DC power flow equation constraint, the transient stability inequality constraint, the line active power flow upper and lower limit safety constraints, and the feasibility inequality constraint of the active output adjustment of the key generator set. The expression of the optimization decision model is as follows: ； ； in, and are respectively the positive adjustment amount and negative adjustment amount of the active power of the key generating set, and are respectively the positive adjustment amount of the active power of the key generator set i and the negative adjustment amount of the active power of the key generator set j, is the active power of the key generator set in the initial operation mode, and are the unit cost coefficients for positive and negative adjustment of the active power of the generator set, and They represent the sets of generators participating in the positive and negative regulation of active power, respectively. is the grid node admittance matrix, and is the voltage phase vector of each node, represents a deep cross network model, and Both represent the sensitivity of the active power of the generator set to the critical removal time. Indicates the time when the relay protection device operates and cuts off the fault line. and They represent the upper and lower limits of the branch active power flow respectively. represents the set of power grid branches, Indicates that there is a branch between node i and node j, and They represent the voltage phases of node i and node j respectively, It represents the maximum limit of the positive adjustment of the active power of generator set i, Indicates the maximum limit of negative adjustment of active power of generator set j.

6. A method for preventing and controlling transient stability of a power system based on a deep cross network according to claim 5, characterized in that: In S6, the specific steps of using the successive linear approximation method to solve the optimization decision model to obtain a power grid operation mode that meets the stable operation requirements are: S6.

1. Initialize the parameters of the optimization decision model to determine the active power of the key generator set under the initial operation mode. And set the initial operation mode as the current operation mode; S6.

2. using a deep cross-network to evaluate the critical removal time of each anticipated fault scenario under the current operation mode of the power grid, and estimating the sensitivity of the critical removal time to the active power of the key generator set; S6.3, Substitute the sensitivity of the active power of the key generator set in the current operation mode into the optimization decision model, solve the optimization decision model to obtain the positive adjustment amount of the active power of the key generator set and the negative adjustment of the active power of the key generator set , correcting the active power of the key generator set and setting the new operation mode as the current operation mode; S6.

4. Use the deep cross network to evaluate the transient stability of the current operation mode of the power grid. If the power system can maintain transient stability for any expected fault scenario, proceed to step S6.5; otherwise, proceed to step S6.2; S6.

5. When the active power adjustment amount of the generator set is less than the preset threshold, the active power adjustment strategy of the key generator set is terminated and output, and a power grid operation mode that meets the stable operation requirements is obtained according to the adjustment strategy; otherwise, return to step S6.2.

Images